Normalizers of Tori

ISSN 1364-0380 (on line) 1465-3060 (printed) 1337

eometry & opology G T G T Volume 9 (2005) 1337–1380 GG T T T G T G T Published: 30 July 2005 G T G T T G T G T G T G T G G G G T T

Normalizers of tori

WG Dwyer CW Wilkerson

Department of Mathematics, University of Notre Dame Notre Dame, Indiana 46556, USA Department of Mathematics, Purdue University West Lafayette, Indiana 47907, USA Email: [email protected], [email protected]

Abstract

We determine the groups which can appear as the normalizer of a maximal torus in a connected 2–compact group. The technique depends on using ideas of Tits to give a novel description of the normalizer of the torus in a connected compact Lie group, and then showing that this description can be extended to the 2–compact case.

AMS Classiﬁcation numbers Primary: 55P35, 55R35 Secondary: 22E10

Keywords: Maximal torus, Weyl group, 2–compact group

Proposed: Haynes Miller Received: 3 February 2004 Seconded: Thomas Goodwillie, Ralph Cohen Revised: 25 July 2005 c eometry & opology ublications G T P 1338 W G Dwyer and C W Wilkerson

1 Introduction

Suppose that G is a connected compact Lie group and that T G is a maximal ⊂ torus, or in other words a maximal connected abelian subgroup. The normalizer NT of T lies in a short exact sequence 1 T NT W 1 (1.1) → → → → in which W is a finite group called the Weyl group of G. In this paper we use ideas of Tits [34] to give a particularly simple description of the groups which appear as such an NT (see Proposition 1.6). This leads to an analogous determination of the groups which appear as the normalizer NT˘ of a maximal 2–discrete torus T˘ in a connected 2–compact group (see Section 1.16). In the connected compact Lie group case, NT determines G up to isomorphism; see [8] for semisimple groups, and [30] or [28] in general. In listing the possible NT s we are thus giving an alternative approach to the classification of connected compact Lie groups. In contrast, it is not known that the normalizer of a maximal 2–discrete torus in a connected 2–compact group X determines X up to equivalence. However, this seems likely to be true [27, 23, 2], and we hope that the results of this paper will eventually contribute to a classification of connected 2–compact groups.

1.2 Compact Lie groups A glance at (1.1) reveals that NT is determined up to isomorphism by three ingredients: (1) the torus T , (2) the finite group W , with its conjugation action on T , and (3) an extension class k H2(W ; T ). ∈ Let T(1) denote the circle R/Z. The torus T is a product T(1)r of circles, and so is determined by the number r, which is the rank of T (or of G). The natural map Aut(T ) Aut(π T ) = GL(r, Z) is an isomorphism, and the → 1 ∼ resulting conjugation homomorphism W GL(r, Z) embeds W as a finite → subgroup of GL(r, Z) generated by reflections. Here a reflection is a matrix which is conjugate in GL(r, Q) to the diagonal matrix diag( 1, 1,..., 1). This − takes care of ingredients (1) and (2). Ingredient (3) is more problematical. The most direct way to approach (3) would be to list all of the elements of H2(W ; T ) and then point to ones which correspond to NT extensions, but it would be a daunting task to make all of these cohomology calculations (one for each finite group acting on a torus by reflections) and then name the resulting elements in a way which would make it convenient to single out the ones giving an NT .

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1339

We take another approach, which is based on [34]. Suppose that T is a torus and W is a finite group of automorphisms of T generated by reflections. A marking for (T,W ) is defined to be a collection h of elements of T , one { σ} for each reflection σ in W , which satisfy some simple conditions (see Defini- tion 2.12). The triple (T, W, h ) is called a marked reflection torus. The { σ } following proposition is less deep than it may seem; it is derived from a simple algebraic correspondence between marked reflection tori and classical root systems (see Section 2).

1.3 Proposition (See Remark 5.6) Suppose that G is a connected compact Lie group with maximal torus T and Weyl group W . Then G determines a natural marking h for the pair (T,W ), and the assignment G (T, W, h ) { σ} 7→ { σ } gives a bijection between isomorphism classes of connected compact Lie groups and isomorphism classes of marked reﬂection tori.

To each such marked reﬂection torus we associate a group ν(T, W, h ), called { σ } the normalizer extension of W by T (see Deﬁnition 3.5), which lies in a short exact sequence 1 T ν(T, W, h ) W 1. (1.4) → → { σ} → → The following result is essentially due to Tits [34].

1.5 Proposition (See Theorem 5.7) Suppose that G is a connected compact Lie group and that (T, W, h ) is the corresponding marked reﬂection { σ } torus. Then the normalizer of T in G is isomorphic to ν(T, W, h ). { σ}

This leads to our characterization of NT s.

1.6 Proposition Suppose that N is an extension of a ﬁnite group W by a torus T . Then N is isomorphic to the normalizer of a torus in a connected compact Lie group if and only if the conjugation action of W on T expresses W as a group of automor- • phisms of T generated by reﬂections, and there exists a marking h of (T,W ) such that N is isomorphic to • { σ} ν(T, W, h ). { σ }

In fact the marking h in Proposition 1.6 is unique, if it exists (see Re- { σ} mark 5.6).

eometry & opology, Volume 9 (2005) G T 1340 W G Dwyer and C W Wilkerson

1.7 2–compact groups A 2–compact group is by definition a pair (X,BX) of spaces together with an equivalence X ΩBX , where H (X; Z/2) is finite ∼ ∗ and BX is 2–complete. Usually we refer to X itself as the 2–compact group and leave BX understood. Any such X has a maximal 2–discrete torus T˘ (see Section 1.16), which is an ordinary discrete group of the form (Z/2∞)r for some r > 0. There is also an associated normalizer NT˘ which lies in a short exact sequence 1 T˘ NT˘ W 1 → → → → in which W is a finite group called the Weyl group of X . The automorphism group of T˘ is isomorphic to GL(r, Z2), and again the conjugation action of W on T˘ embeds W as a subgroup of Aut(T˘) generated by reflections. In this context a reflection is an element of GL(r, Z2) which is conjugate in GL(r, Q2) to diag( 1, 1,..., 1). As above, we define a marking for (T,W˘ ) to be a suitable − (see Definition 6.8) collection of elements h in T˘, and we associate to each { σ} such marking a group ν(T˘ , W, h ), called the normalizer extension of W by { σ} T˘ (see Definition 6.15), which lies in a short exact sequence 1 T˘ ν(T˘ , W, h ) W 1. (1.8) → → { σ } → → The structure (T˘ , W, h ) goes by the name of marked 2–discrete reflection { σ } torus. Our main results are the following ones.

1.9 Proposition (See Lemma 9.9) Suppose that X is a connected 2– compact group with maximal 2–discrete torus T˘ and Weyl group W . Then X determines a natural marking h for (T,W˘ ), and thus a marked 2–discrete { σ} reflection torus (T˘ , W, h ). { σ } 1.10 Proposition (See Proposition 9.12) Suppose that X is a connected 2–compact group and that (T˘ , W, h ) is the corresponding (see Proposi- { σ } tion 1.9) marked 2–discrete reflection torus. Then the normalizer of T˘ in X is isomorphic to ν(T˘ , W, h ). { σ } It turns out that there are not very many marked 2–discrete reflection tori to work with. If (T, W, h ) is a marked reflection torus, there is an associated { σ } marked 2–discrete reflection torus (T δ, W, h ), where { σ } T δ = x T 2kx = 0 for k 0 . (1.11) { ∈ | ≫ } Say that a 2–discrete marked reflection torus is of Coxeter type if it is derived in this way from an ordinary marked reflection torus. Let denote the marked T∆ 2–discrete reflection torus derived from the exceptional 2–compact group DI(4) [10, 24]. There is an evident notion of cartesian product for marked 2–discrete reflection tori (see Section 7).

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1341

1.12 Proposition (See Propositions 6.10 and 7.4) Any marked 2–discrete reﬂection torus can be written as a product , where is of Coxeter T1 ×T2 T1 type and is a product of copies of . T2 T∆

It follows from Proposition 1.12 that any marked 2–discrete reﬂection torus arises from some 2–compact group: those of Coxeter type from 2–completions of connected compact Lie groups (Section 8 and Lemma 9.17), and the others from products of copies of DI(4). This can be used to give a characterization parallel to Proposition 1.6 of normalizers of maximal 2–discrete tori in connected 2–compact groups.

1.13 Remark Proposition 1.12 is in line with the conjecture that any connected 2–compact group X can be written as X X , where X is the 2– 1 × 2 1 completion of a connected compact Lie group and X2 is a product of copies of DI(4) (see Proposition 9.13). Taken together, Propositions 1.10 and 1.12 imply that the conjecture is true as far as normalizers of maximal 2–discrete tori are concerned.

1.14 Remark If p is an odd prime, the normalizer of a maximal p–discrete torus in a connected p–compact group is always the semidirect product of the torus and the Weyl group [1]. This explains why we concentrate on p = 2; for odd p the extension we study is trivial. The techniques of this paper work to some extent for odd primes and can be used to show that the extension is trivial when the Weyl group is a Coxeter group; most of the remaining cases can be handled with the general methods of [15].

1.15 Organization of the paper Section 2 describes various ways of inter- preting a root system, one of which is in terms of a marked reflection torus. Section 3 shows how a marked reflection torus (T, W, h ) gives rise to a nor- { σ } malizer extension ν(W ) of W by T ; the normalizer extension is defined with the help of a reflection extension ρ(W ) of W . In Section 4 we find generators and relations for ρ(W ) by identifying this group with an extension τ(W ) of W constructed by Tits. Section 5 exploits these generators and relations to show that the normalizer NT of a maximal torus in a connected compact Lie group G is isomorphic to the normalizer extension obtained from the marked reflection torus (equivalently root system) associated to G. The arguments are due to Tits [34], although we put them in a different context and present them in a way that can be generalized to 2–compact groups. Section 6 generalizes the material from Sections 2 and 3 to a 2–primary setting, and Section 7 goes on to

eometry & opology, Volume 9 (2005) G T 1342 W G Dwyer and C W Wilkerson give a classiﬁcation of marked 2–discrete reﬂection tori. Section 8 gathers together some technical information about 2–completions of connected compact Lie groups. Finally, Section 9 proves the main results about the normalizer of a maximal 2–discrete torus in a connected 2–compact group; the approach depends in part on a technical lemma, which is proved in Section 10.

1.16 Notation and terminology By a maximal 2–discrete torus T˘ in a 2– compact group X , we mean the discrete approximation [11, 6.4] to a maximal 2–compact torus T in X [11, 8.9] [12, 2.15]. The normalizer of T˘ is the discrete approximation in the sense of [12, 3.12] to the normalizer of T [11, 9.8] (Aside 9.1). We takeb for granted the basic properties of 2–compact groups (see for instance the survey articles [20, 22, 26]), although we sometimesb recall the definitions to help orient the reader. Since we only work with p–compact groups for p = 2, we sometimes abbreviate 2–complete to complete and 2–discrete to discrete. Given an abelian group A and an involution σ on A, we write A−(σ) for the kernel of 1 + σ on A, and A+(σ) the kernel of 1 σ. − The following equations give the relationship between our notions of root system (Definition 2.2) and geometric root system (see Section 2.23) and similar notions of Tits, Serre, and Bourbaki. root system = système de racines [34] = root data [32] = diagramme radiciel réduit [5, 4, Sect. 8] § geometric root system = système de racines [31] [4] In a nutshell, for us root systems are integral objects, while geometric root systems are real objects. It is a classical result [4] that geometric root systems classify connected compact semisimple Lie groups up to isogeny; equivalently, they classify simply connected compact Lie groups (or center–free connected compact Lie groups) up to isomorphism. A root system (in our sense) is an integral analog of a geometric root system, an analog which has the additional feature of allowing for the existence of central tori; root systems classify ar- bitrary connected compact Lie groups up to isomorphism. The existence of a natural correspondence between root systems and connected compact Lie groups is documented in [5, 4, Sect. 9]. This correspondence can be derived § fairly easily from the classical correspondence between simple geometric root systems and simply connected simple compact Lie groups. However, there is an indirect path to the correspondence that goes through the intermediate category of connected reductive linear algebraic groups over C (see Section 2.23).

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1343

The authors would like to thank Kasper Andersen, Jesper Grodal, and the referee for their comments and suggestions; they are particularly grateful to T A Springer for pointing out some references to the literature.

2 Root systems, lattices and tori

In this section we show that three different structures are equivalent: root systems, marked reflection lattices, and marked reflection tori. The conclusion is that a root system amounts to a reflection lattice L together with a bit of extra marking data for each reflection; this marking data is particularly easy to specify in terms of the torus T(1) L. ⊗ 2.1 Root systems The following definition is adapted from [34, 4.1]. A lattice L (over Z) is a free abelian group of finite rank.

2.2 Definition A root system in a lattice L consists of a finite subset R of L (the set of roots) and for each r R a homomorphism n : L Z (the coroot ∈ r → associated to r) such that following axioms are satisfied: (R1) taken together, R and ker(n ) span Q L; r r ⊗ (R2) for each r R, n (r)= 2; ∈ r T− (R3) if r R, k Z, and kr R, then k = 1; ∈ ∈ ∈ ± (R4) if r, t R, then t + n (t)r R. ∈ r ∈ 2.3 Remark This structure is called a système de racines in [34] and a diagramme radiciel rèduit in [5]. It is not the conventional système de racines of Serre [31, Section 5] or Bourbaki [4, VI], which we will call a geometric root system. Basically, a root system in our sense is an integral form of a geometric root system; see Section 2.23 below.

2.4 Marked Reflection Lattices A reflection on a lattice L is an automorphism σ such that σ is conjugate in Aut(Q L) = GL(r, Q) to a diagonal ⊗ ∼ matrix diag( 1, 1,..., 1). − 2.5 Definition Suppose that σ is a reflection on a lattice L. A strict marking for σ is a pair (b, β), where b L and β : L Z is a homomorphism such ∈ → that for any x L ∈ σ(x)= x + β(x)b. (2.6) Two strict markings (b, β) and (b′, β′) are equivalent if (b, β) = (b′, β′). A ± marking for σ is an equivalence class of strict markings.

eometry & opology, Volume 9 (2005) G T 1344 W G Dwyer and C W Wilkerson

Suppose that L is a lattice, σ is a reflection on L with marking (b, β), and ± w is an automorphism of L. We let w (b, β) denote the marking for wσw−1 · given by (w(b), β w−1). ◦ 2.7 Definition A reflection lattice is a lattice L together with a finite subgroup W of Aut(L) which is generated by the reflections it contains. A marked reflection lattice is a reflection lattice (L,W ) together with markings (b , β ), ± σ σ one for each reflection σ in W , such that for w W , ∈ w (b , β )= (b −1 , β −1 ). · σ σ ± wσw wσw 2.8 Marked reflection tori Let T(n) denote the n–torus (R/Z)n . A torus is a compact Lie group isomorphic to T(n) for some n. Any torus T gives a lattice π T ; conversely, a lattice L gives a torus T(1) L. These two con- 1 ⊗ structions are inverse to one another up to natural isomorphism, and induce an equivalence between the category of tori (with continuous homomorphisms) and the category of lattices.

2.9 Remark We write the group operation in a torus additively. If T is a torus, the lattice π1T can also be described as the group Hom(T(1), T ) of continuous homomorphisms from the 1–torus to T (these are cocharacters of T ).

2.10 Definition An automorphism σ of T is said to be a reflection (respectively, trivial mod 2) if the induced automorphism of π1T is a reflection (respectively, trivial mod 2). An element x of T is said to be strongly σ–negative − − if x lies in the identity component T0 (σ) of T (σ) (see Section 1.16), or, in other words, if x lies in the identity component of the group T −(σ)= x T σ(x)= x . { ∈ | − } 2.11 Remark A reflection σ on T is trivial mod 2 if and only if σ acts as the identity on T = x T 2x = 0 . If T = T(1) L, then x T is strongly 2 { ∈ | } ⊗ ∈ σ–negative if and only if x T(1) L−(σ). ∈ ⊗ 2.12 Definition Suppose that σ is a reflection on a torus T . A marking for σ is an element h T such that ∈ (1) h is strongly σ–negative (Remark 2.11), (2) 2h = 0, and (3) h =0 if σ is nontrivial mod 2. 6

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1345

2.13 Definition A reflection torus is a torus T together with a finite subgroup W of Aut(T ) which is generated by the reflections it contains. A marked reflection torus is a reflection torus (T,W ) together with markings h T , σ ∈ one for each reflection σ in W , such that for w W , ∈

hwσw−1 = w(hσ ).

2.14 Remark If (L,W ) is a reflection lattice, or (T,W ) is a reflection torus, then W is a classical Weyl group [19, 2.9] and hence a Coxeter group [19, Section 1]. Let S = s ,...,s be a set of simple reflections in W [19, 1.3], { 1 ℓ} and mi,j the order of sisj , so that mi,j is the Coxeter matrix associated to S . { } mi,j By [19, 1.9], W is generated by S subject only to the relations (sisj) = 1.

2.15 Equivalence between the three structures The goal of this section is to prove the following two statements.

2.16 Proposition If L is a lattice, then there is a natural bijection between marked reﬂection structures on L and root systems in L.

2.17 Proposition If L is a lattice with associated torus T = T(1) L, then ⊗ there is a natural bijection between marked reﬂection structures on L and marked reﬂection structures on T .

2.18 Remark It is not hard to enumerate the possible markings for a reflection σ on L. Write ker(1+σ) and im(1 σ) for the kernel and image of the indicated − endomorphisms of L; since σ is a reflection, both of these groups are infinite cyclic. If b L is an element such that im(1 σ) b , then for every x L ∈ − ⊂ h i ∈ there is a unique integer β(x) such that σ(x) = x + β(x)b; it is easy to check that β : L Z is a homomorphism, and so (b, β) is a strict marking. In → particular, strict markings for σ correspond bijectively to such elements b, and markings to such elements b taken up to sign. Any such b lies in ker(1 + σ). It is easy to see that there are inclusions

2 ker(1 + σ) im(1 σ) ker(1 + σ). ⊂ − ⊂ where the left hand inclusion is an equality if and only if σ reduces to the identity mod 2. Let b0 be a generator of ker(1 + σ). Then σ has exactly two markings (determined by b and 2b ) if σ is trivial mod 2, and one marking ± 0 ± 0 (determined by b ) otherwise. ± 0

eometry & opology, Volume 9 (2005) G T 1346 W G Dwyer and C W Wilkerson

Proof of Proposition 2.16 Suppose to begin with that (L,W ) is a marked reflection lattice. For each reflection σ W , let (b , β ) be the corresponding ∈ ± σ σ marking, so that σ(x)= x + βσ(x)bσ. (2.19) The set of roots in the corresponding root system is given by R = b σ W a reflection . {± σ | ∈ } For each root r R, the coroot n is β if r = b and β if r = b . It ∈ r σ σ − σ − σ remains to check axioms (R1)-(R4). For (R1), note that r ker(nr) is the fixed point set LW of the action of W on L, and observe that for any x L there T ∈ is an expression (#W )x = w(x)+ (x w(x)), − wX∈W wX∈W with the first summand on the right in LW and the second in the span of R. Condition (R2) comes from combining (2.19) with the equality σ2(x)= x. For each w W and reflection σ W , w(b )= b −1 (Definition 2.7); applied ∈ ∈ σ ± wσw to the special case in which w is another reflection, this gives (R4). For (R3) we take the following argument from [31, V.1]. Suppose that σ and τ are reflections in W such that both bσ and bτ are multiples of a single element b L; we must show that σ = τ . Let w = στ . Then w(b)= b, and w acts as ∈ the identity on Q (L/ b ); in particular, all of the eigenvalues of w are equal ⊗ h i to 1. But w W has finite order, and so w must be the identity map. ∈ On the other hand, suppose that R is a root system in L. For each root r let σ : L L be given by r → σr(x)= x + nr(x)r. It follows from (R2) that σ is a reflection on L, with a marking (r, n ); take r ± r W to be the group generated the transformations σr . By (R4) each reflection σr preserves the finite set R of roots, and so W also preserves R. To see that W is finite, it is enough to check that any element of W which acts trivially on R is the identity, but this follows from (R1) and the fact that the subspace r ker(nr) is pointwise fixed by each reflection σr and hence pointwise fixed by W . Note −1 T that by the above eigenvalue argument, wσrw = σw(r) : both transformations send w(r) to w(r) and act as the identity on Q (L/ w(r) ). Checking that − ⊗ h i the markings are preserved under conjugation thus amounts to showing that if σ = σ then r = t (note that as in Remark 2.18 a marking (r, n ) is r t ± ± r determined by its first component). But since r and t then determine markings of the same reflection, it follows from Remark 2.18 that r = t, r = 2t, or ± ± t = 2r, and the last two possibilities are ruled out by (R3). The one remaining ±

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1347 issue is that there might be reflections in W which do not appear in the set σ r R ; these reflections have not been marked in the above discussion. { r | ∈ } But by [19, 1.14] there are no such additional reflections.

Proposition 2.17 is a consequence of the following lemma, which is proved by examining the discussion in Remark 2.18.

2.20 Lemma Let σ be a reflection on the lattice L, treated also as a reflection on T = T(1) L = (R L)/L. Then sending b L to the residue class of ⊗ ⊗ ∈ b/2 in T gives a bijection between markings (b, β) of σ and elements h T ± ∈ satisfying the conditions of Definition 2.12.

2.21 Duality A reflection σ on L induces a reflection on the dual lattice L# = Hom(L, Z); sending (b, β) to (β, b) gives a bijection between markings ± ± of σ and markings of the dual reflection. It follows that if (L,W ) is a marked reflection lattice with markings (b , β ), then (L#,W ) is a marked reflection ± σ σ lattice with markings (β , b ). Similarly, interchanging roots and coroots give ± σ σ a bijection between root systems in L and root systems in L# .

2.22 Counting root systems As in Remark 2.18, the number of markings of a reflection lattice (L,W ) is 2k , where k is the number of conjugacy classes of reflections in W which are trivial mod 2. By Proposition 2.16, this is also the number of distinct root systems with reflection lattice (L,W ).

2.23 Another notion of root system We point out the relationship between our notion of root system (which is Tits’) and that of Serre and Bourbaki. Recall that what we term a geometric root system (Remark 2.3) is a pair (V, R) such that V is a finite-dimensional real vector space, R is a finite subset of V , and the following conditions are satisfied [31, Section V]. R generates V as a vector space, and does not contain 0. • For each r R there exists a reflection s on V which carries R to itself • ∈ r and has s (r)= r. r − For each r, t R, s (t) t is an integral multiple of r. • ∈ r − The geometric root system is said to be reduced if for each r R, r and r ∈ − are the only roots proportional to r. The following proposition is immediate. We will say that a root system (see Definition 2.2) is semisimple if ker(n )= 0 . r r { } T eometry & opology, Volume 9 (2005) G T 1348 W G Dwyer and C W Wilkerson

2.24 Proposition If (L,R, n ) is a semisimple root system, then (R L, R) { r} ⊗ is a a reduced geometric root system.

Conversely, if (V, R) is a reduced geometric root system, then for each r R ∈ deﬁne a homomorphism H : V R by the formula s (x) = x + H (x)r. Let r → r r L be the lattice in V given by x r, H (x) Z , and let L be the max { | ∀ r ∈ } min lattice in L spanned by R. Note that if L V is a lattice between L and ⊂ min L , then H (L) Z. max r ⊂ 2.25 Proposition Suppose that (V, R) is a reduced geometric root system, and that L is a lattice in V with Lmin L Lmax . Then (L,R, H ) is a ⊂ ⊂ { r} semisimple root system.

Recall that a connected compact Lie group G is said to be semisimple if π1G is finite, or equivalently if the center of G is finite. It is well–known [4] that geometric root systems classify connected semisimple compact Lie groups up to isogeny, ie up to finite covers. A glance at Serre’s description of the fundamental group of a connected compact Lie group [31, VIII-10] shows that extra lattice data in a semisimple root system (Proposition 2.25) allow such systems to classify connected semisimple compact Lie groups up to isomorphism. More generally, root systems classify connected compact Lie groups up to isomorphism: the lattice r nr in the root system derived from a connected compact Lie group G determines the torus which is the identity component of the T center of G. In our way of assigning a root system to G (see Section 5), which is dual to the ordinary one (see Remark 5.6), r nr is the fundamental group of the central torus. This classification theorem is stated in [5, 8, Sect. 9]. T § It can also be obtained by combining the correspondence between compact Lie groups and reductive complex algebraic groups [29, p247] with the classification of connected reductive complex algebraic groups by root data [32]. Springer’s root data are essential the same as our root systems.

3 The normalizer and reﬂection extensions

In this section we take a marked reflection torus (T, W, h ) and construct { σ } the associated normalizer extension of W by T . Let Σ W be the set of ⊂ reflections in W ; the group W acts on Σ by conjugation, and hence on the free abelian group Z[Σ] generated by Σ. We first describe an extension of W by Z[Σ], called the reflection extension of W , which depends only on the structure

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1349 of W as a reﬂection group, and then use this in conjunction with the markings h to obtain the normalizer extension. { σ}

3.1 The reflection extension Write Σ = i Σi as a union of conjugacy classes of reflections. For each index i choose a reflection τ in Σ , and let C ` i i i be the centralizer of τ in W . Let L = π T , and let a L be a nonzero i 1 i ∈ element with τ (a ) = a . It is clear that each element of C takes a to i i − i i i a , and that C is isomorphic to τ C⊥ , where C⊥ is the subgroup of C ± i i h ii× i i i consisting of elements which fix ai . As usual, extensions of W by Z[Σ] are classified by elements of H2(W ; Z[Σ]). There are isomorphisms H2(W ; Z[Σ]) = H2(W ; Z[Σ ]) = H2(C ; Z) ∼ ⊕i i ∼ ⊕i i where the last isomorphism comes from Shapiro’s lemma. Thus in order to specify an extension of W by Z[Σ] it is enough to specify for each i an extension of Ci by Z. We do this as follows. For each i write C = τ C⊥ = Z/2 C⊥ i ∼ h ii× i ∼ × i and consider the obvious extension 1 Z Z C⊥ Z/2 C⊥ 1. (3.2) → → × i → × i → 2 This extension corresponds to the element of H (Ci; Z) which is a pullback under the projection C Z/2 of the unique nonzero element in H2(Z/2; Z). i → By Shapiro’s lemma there is a corresponding extension ρi(W ) of W by Z[Σi], and it is easy to see that up to canonical isomorphism this extension of W does not depend on the choice of representative τi for the conjugacy class Σi .

3.3 Definition The reflection extension ρ(W ) of W is the extension of W by Z[Σ] determined by the above collection of extensions of the groups Ci ; in other words, ρ(W ) is the fibrewise product (over W ) of the extensions 1 Z[Σ ] ρ (W ) W 1. → i → i → → 3.4 The normalizer extension Given a marked reflection torus = (T, W, h ), T { σ } there is a W –map f T : Z[Σ] T → which sends a reflection σ to hσ .

eometry & opology, Volume 9 (2005) G T 1350 W G Dwyer and C W Wilkerson

3.5 Definition Suppose that = (T, W, h ) is a marked reflection torus. T { σ } The normalizer extension ν(T, W, h ) of W by T is defined to be the image { σ } of the reflection extension ρ(W ) under the map f T : Z[Σ] T . → 3.6 Remark The reflection extension of W is determined by an extension class k H2(W ; Z[Σ]), and the normalizer extension by f T (k) H2(W ; T ). ∈ ∗ ∈

4 The Tits extension

Suppose that (L,W ) is a reﬂection lattice. In this section we derive generators and relations (Section 4.6) for the reﬂection extension ρ(W ) of W . To accom- plish this, we identify ρ(W ) with the extended Weyl group constructed by Tits [34]; we call this extended Weyl group the Tits extension of W and denote it τ(W ). The presentation for ρ(W ) is used both in the proof of Theorem 5.7 and later on in the proof of Proposition 9.12 (Coxeter case).

4.1 The Tits extension Let S = s ,...,s be a chosen set of simple { 1 ℓ} reﬂections in W (see Remark 2.14), and let Σ∗ be a copy of the set Σ of all reﬂections in W . As in Section 3, W acts by conjugation on Σ and thus on Z[Σ] or Z[Σ∗].

4.2 Deﬁnition The Tits extension τ(W ) of W is the subgroup of the semidirect product Z[Σ∗] ⋊ W generated by the elements (s∗,s), s S . ∈ Since the set S generates W [19, 1.5], the natural surjection Z[Σ∗] ⋊ W W → restricts to a surjection τ(W ) W . The kernel of this map is contained in → Z[Σ∗], and Tits [34, Cor. 2.7] shows that it is exactly 2Z[Σ∗], which of course is isomorphic to Z[Σ] as a W –module. Under this identiﬁcation, we can treat τ(W ) as an extension 1 Z[Σ] τ(W ) W 1 (4.3) → → → → of W by Z[Σ]. In this section we will prove the following proposition.

4.4 Proposition Suppose that (L,W ) is a reflection lattice and that Σ is the set of reflections in W . Then, as an extension of W by Z[Σ], the reflection extension ρ(W ) is isomorphic to the Tits extension τ(W ).

4.5 Remark It follows from Proposition 4.4 that the isomorphism class of the extension τ(W ) is independent of the choice of the set S of simple reﬂec- tions. This also follows from the fact that any two sets of simple reﬂections are conjugate in W [19, 1.4].

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1351

4.6 Generators and relations Let (mi,j) be the Coxeter matrix (see Re- mark 2.14) of (L,W ), so that W is generated by simple reﬂections s ,...,s { 1 ℓ} subject to the relations mi,j (sisj) = 1. (4.7)

Note that mi,i = 1. Following Tits [34, 0.1], we put these relations in a slightly different form. Write prod(n; y,x)= yxyx · · · where the product on the right contains exactly n factors, alternating between x and y, and beginning on the right with an x. Then W is generated by the elements s , 1 i ℓ, subject only to the relations i ≤ ≤ 2 si = 1 prod(m ; s ,s ) = prod(m ; s ,s ) for i = j. i,j i j i,j j i 6 −1 In fact the first relation gives si = si , while the second one is obtained from (4.7) by the reversible process of moving the leading terms of the expanded mi,j expression (sisj) to the right hand side of the equation and replacing each −1 occurrence of x (x = si or sj ) by x itself. In [34, 2.6], Tits gives a presentation for τ(W ) parallel to the above presentation for W . For notational reasons, think for the moment of τ(W ) as an extension ∗ ∗ of W by Z[Σ ]. Then τ(W ) is generated by Z[Σ ] and symbols qi , one for each simple reflection, subject to the following relations: 2 ∗ qi = si q σ∗q−1 = (s σs−1)∗, σ Σ i i i i ∈ prod(m ; q ,q ) = prod(m ; q ,q ) for i = j. i,j i j i,j j i 6 With respect to this presentation of τ(W ), the quotient map τ(W ) W sends → qi to si . By Proposition 4.4, these same formulas give a presentation of ρ(W ).

The rest of this section is devoted to the proof of Proposition 4.4. We ﬁrst need an explicit formulation of Shapiro’s lemma. As above, if G is a ﬁnite group and H G is a subgroup, extensions of G by the permutation module Z[G/H] ⊂ correspond bijectively to extensions of H by the trivial H –module Z. We will call these two extensions Shapiro companions of one another. The following lemma is elementary; see [6, III.8, Ex. 2].

4.8 Lemma Suppose that G is a ﬁnite group, H G a subgroup, and ⊂ 1 Z[G/H] E G 1 (4.9) → → → →

eometry & opology, Volume 9 (2005) G T 1352 W G Dwyer and C W Wilkerson an extension of G by the permutation module Z[G/H]. Then the Shapiro companion of (4.9) is obtained by pulling the extension back over H to obtain 1 Z[G/H] E′ H 1, → → → → and then taking the quotient of E′ by the subgroup of Z[G/H] generated by the elements x G/H, x = eH . { ∈ 6 }

4.10 Remark Note that H ﬁxes the coset eH in G/H , and so H carries the set x G/H, x = eH to itself. This implies that the subgroup of Z[G/H] { ∈ 6 } generated by this set is a normal subgroup of the group E′ above. The quotient Z[G/H] / Z[x G/H, x = eH] ∈ 6 is canonically isomorphic to the trivial H –module Z, with generator given by the residue class of eH .

We next recall some apparatus from [34]. Let I be a set of indices corresponding to the chosen simple reflections in W , and I the free group on I . An element i I is said to be positive if it is a product of elements of I ; the length l(i) is ∈ then the number of elements in this product expression. To conform with [34], for each i I we will denote the corresponding simple reflection in W by r ∈ i (instead of s ); there is a unique (surjective) homomorphism r : I W with i → r(i)= r , i I . A positive word i I is said to be minimal if there does not i ∈ ∈ exist a positive word j I with l(j) < l(i) and r(j) = r(i). If w W , the ∈ ∈ length of w is defined to be the length of a minimal positive word in I which maps to w under r. Suppose that i = i i i is a positive word in I of length m. For each 1 2 · · · m integer k with 1 k m set ≤ ≤ σ = r(i i i ) r r(i i i )−1. k 1 2 · · · k−1 · ik · 1 2 · · · k−1 Each one of these elements is a reflection in W , and the sequence σ ,σ , ,σ 1 2 · · · m is said to be the sequence of reflections associated to the word i. Recall the following proposition.

4.11 Proposition [34, 1.6] Let i I be a minimal word, σ W a reﬂection, ∈ ∈ and n the number of times which σ appears in the sequence of reﬂections associated to i. Then n = 0 or 1.

We need to evaluate the integer n above in a special case.

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1353

4.12 Proposition Let i I be a minimal word, σ W a reflection, and n the ∈ ∈ number of times which σ appears in the sequence of reflections associated to i. Suppose that σ is a simple reflection, that r(i) is a reflection which commutes with σ, and that r(i) = σ. Then n = 0. 6 The proof of this was explained to us by M. Dyer. It depends on the following lemma.

4.13 Lemma [17, 1.4] Suppose that t = r(i i i ) is a reﬂection in W 1 2 · · · 2n+1 with l(t) = 2n + 1. Then t = r(i i i i i ). 1 · · · n n+1 n · · · 1 For the convenience of the reader, we give the argument from [17].

Proof of Lemma 4.13 Let x = r(in i1) and y = r(in+2 i2n+1). Then · · · · · · −1 l(x) = l(y) = n and l(rin+1 x) = l(rin+1 y) = n + 1. Note that t = t , so −1 that rin+1 yt = rin+1 yt = x and l(rin+1 yt) < l(rin+1 y). Applying the Strong Exchange Condition [19, 5.8] to this last observation shows that there is a number k between n + 1 and 2n + 1 such that x = r yt = r(i i i ). in+1 n+1 · · · k · · · 2n+1 Since l(x) = n, this is a minimal expression for x. Since l(r x) > l(x) it b in+1 must be the case that k = n + 1 and hence that x = y.

Proof of Proposition 4.12 Write i = i i i , and suppose that σ ap- 1 2 · · · m pears among the sequence σ ,σ , ,σ of reﬂections associated to i, that is, 1 2 · · · m suppose that there exists k with 1 k m such that ≤ ≤ σ = σ = r(i i )r r(i i )−1. k 1 · · · k−1 ik 1 · · · k−1 Let t = r(i i ); it is clear that 1 · · · m t = r(i i ) r r(i i )−1r(i i )r(i i ) 1 · · · k−1 · ik · 1 · · · k−1 1 · · · k−1 k+1 · · · m = σ r(i i )r(i i ). · 1 · · · k−1 k+1 · · · m Choose i I so that σ = r (recall that σ is a simple reﬂection) and let 0 ∈ i0 j = i i i i i ; then r(j) = t and, since l(j) = l(i), j is a minimal 0 1 · · · k−1 k+1 · · · m word with r(j) = t. Write j = j j , with j = i , so that r = σ. For 1 · · · m 1 0 j1 determinant reasons m = 2q +1 is odd, and m> 1 because t = r(j) = σ. By 6 Lemma 4.13, t = r r(j j j j j ) r j1 · 2 · · · q q+1 q · · · 2 · j1 = σ r(j j j j j ) σ. · 2 · · · q q+1 q · · · 2 ·

eometry & opology, Volume 9 (2005) G T 1354 W G Dwyer and C W Wilkerson

Now consider σ t σ. Since t commutes with σ and σ2 = 1, this element is · · equal to t. In combination with the above equation, we get t = σ t σ · · = σ σ r(j j j j j ) σ σ · · 2 · · · q q+1 q · · · 2 · · = r(j j j j j ), 2 · · · q q+1 q · · · 2 which, in contradiction to the assumption that i is a minimal word, gives l(t) ≤ m 2. This contradiction shows that σ does not appear in the list of reﬂections − associated to i.

Finally, one more calculation from Tits. Let a: I Z[S∗] ⋊ W be the homo- → morphism determined by a(i) = (r∗, r ), i I . By deﬁnition, the image of a is i i ∈ τ(W ).

4.14 Lemma [34, Lemme 2.4] Let i I be a positive word of length m, and ∈ σ ,σ , ,σ the sequence of reﬂections associated to i. Then 1 2 · · · m m ∗ a(i)= σk, r(i) . ! Xk=1

Proof of Proposition 4.4 We identify the kernel of the map τ(W ) W → with Z[Σ]. Then τ(W ) sits inside the semidirect product Z[Σ∗]⋊W , and there is map of short exact sequences 1 Z[Σ] τ(W ) W 1 −−−−→ −−−−→ −−−−→ −−−−→

 ∗ ∗  1 Z[Σ ] Z[Σ ] ⋊ W W 1 −−−−→ y −−−−→ y −−−−→ y −−−−→ in which the left hand vertical map gives an isomorphism between Z[Σ] and 2Z[Σ∗]. As in Section 3.1, giving the extension τ(W ) amounts to giving a collection of extensions τ (W ) , where τ (W ) is an extension of W by Z[Σ ]; { i } i i the extension τi(W ) is obtained from τ(W ) by taking a quotient of τ(W ) by the subgroup of Z[Σ] generated by Σ Σi . Clearly τi(W ) lies inside the semidirect ∗ \ product Z[Σi ] ⋊ W , and in fact there is a map of short exact sequences 1 Z[Σ ] τ (W ) W 1 −−−−→ i −−−−→ i −−−−→ −−−−→ (4.15)  ∗ ∗  1 Z[Σ i ] Z[Σi] ⋊ W W 1 −−−−→ y −−−−→ y −−−−→ y −−−−→

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1355

∗ in which as before the left hand vertical map identifies Z[Σi] with 2Z[Σi ]. Pulling the diagram (4.15) back over the inclusion C W gives a commutative i → diagram: 1 Z[Σ ] E C 1 −−−−→ i −−−−→ i −−−−→ i −−−−→ (4.16)  ∗ ∗  1 Z[Σ i ] Z[Σi] ⋊ Ci Ci 1 −−−−→ y −−−−→ y −−−−→ y −−−−→ Now taking quotients as in Lemma 4.8, in the upper group by Z[Σ τ ] and in i \ i the lower by Z[Σ∗ τ ∗], yields: i \ i 1 Zτ E′ C 1 −−−−→ i −−−−→ i −−−−→ i −−−−→ (4.17) ∗ ∗  1 Zτi Zτi  Ci Ci 1 −−−−→ y −−−−→ y× −−−−→ y −−−−→∗ It is clear that the left hand vertical map sends τi to 2τi . We need to argue that the upper extension is the one described in (3.2). Note that every reflection in W is conjugate to a simple reflection [19, 1.14], so we can assume that τi belongs to the set S of simple reflections. Recall ⊥ ⊥ that Ci ∼= τi Ci , where Ci can be identified as the subgroup of W which h i× − ⊥ pointwise fixes L (τi). Following Steinberg [33, 1.5] [19, 1.12], Ci is generated by the reflections it contains, and so in particular it is generated by reflections t which commute with τi and are distinct from τi . Pick such a t, and write t = r(i), where i is a minimal positive word. By Proposition 4.12, τi does not appear among the list of reflections associated to i. Clearly the image α of a(i) ′ in τi(W ) belongs to Ei (see (4.16)); let a(i) be the image of α in Ei (see (4.17)). It now follows from Lemma 4.14 that the image of a(i) in Zτ ∗ C is (0,t). i × i The subgroup of Zτ ∗ C generated by these elements as t varies through the i × i reflections in C⊥ is 0 C⊥ . Consider the subgroup E′′ of Zτ ∗ C generated i { }× i i i × i by 0 C⊥ and by (τ ∗,τ ). Clearly, E′′ is contained inside the image of E′ { }× i i i i i in Zτ ∗ C ; on the other hand, it is also easy to check that there is a short i × i exact sequence 1 2Zτ ∗ E′′ C 1. → i → i → i → ′′ ′ ′′ By the five lemma, Ei ∼= Ei , and it is now simple to verify that Ei is the extension of Ci described in (3.2).

5 Compact Lie Groups

In this section, we follow the lead of Tits [34] and describe the normalizer of the torus in a connected compact Lie group in terms of a marked reﬂection torus

eometry & opology, Volume 9 (2005) G T 1356 W G Dwyer and C W Wilkerson

(Deﬁnition 2.12) associated to the group.

Suppose that G is a connected compact Lie group of rank r with maximal torus T , maximal torus normalizer N = NT and Weyl group W = N/T . Then T ∼= T(r) and the conjugation action of W on T identifies W with a subgroup of Aut(T ) generated by reflections [18, VII.2.13] [14, 5.16]. In particular, (T,W ) is a reflection torus.

For each reﬂection σ W we will construct a subgroup N of N , called the ∈ σ root subgroup of N associated to σ. We will also single out a speciﬁc element hσ in T , the marking associated to σ.

+ Given σ, let T0 (σ) denote the identity component of the ﬁxed point set of the action of σ on T . It is easy to see that T +(σ) is a torus of rank r 1; if L 0 − is the lattice corresponding to T , then under the isomorphism T(1) L = T , ⊗ ∼ T +(σ) is given by T(1) L+(σ). Let G(σ) denote the centralizer of T +(σ) in 0 ⊗ 0 G. Then G(σ) is a connected [18, VII.2.8] [14, 7.3] compact Lie subgroup of G with maximal torus T and Weyl group given by the subgroup 1,σ of W . { } + The normalizer of T in G(σ) can be identiﬁed with the centralizer of T0 (σ) in N ; we denote this group by N(σ). The only simply-connected compact Lie group of rank 1 is SU(2); the center of SU(2) is Z/2 and the adjoint form is SO(3). Given this, it follows from the product splitting theorem for compact Lie groups [14, Section 10] that G(σ) is isomorphic to one of the following three groups:

(1) A product SU(2) T(r 1) × − (2) A product SO(3) T(r 1) × − (3) A product U(2) T(r 2) × − − − Recall that T0 (σ) denotes the identity component of T (σ); this group is a torus of rank 1.

5.1 Deﬁnition The root subgroup Nσ is given by N = x N(σ) x is conjugate in G(σ) to some y T −(σ) . σ { ∈ | ∈ 0 }

5.2 Lemma The set Nσ is a closed subgroup of N(σ) with two components. The identity component of N is the circle T −(σ) = N T , and the natural σ 0 σ ∩ map N /T −(σ) N(σ)/T = Z/2 is an isomorphism. If x,y belong to the σ 0 → ∼ nonidentity component of N , then x2 = y2 T −(σ). σ ∈ 0

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1357

Proof The simplest way to prove this is to examine the above three cases. In − the ﬁrst two cases Nσ is the normalizer of the maximal torus T0 (σ) in SU(2) (respectively, SO(3)). In the third case, Nσ is the normalizer of the maximal − torus T0 (σ) of the subgroup SU(2) of U(2).

5.3 Deﬁnition The marking h T −(σ) is the image of the nonidentity σ ∈ 0 component of N under the squaring map x x2 . σ 7→ 5.4 Lemma The element h T −(σ) T is a marking for σ in the sense of σ ∈ 0 ⊂ Deﬁnition 2.12.

Proof Again, this is most easily proved by inspection. The element hσ is strongly σ–negative by construction, and it is easy to see that 2hσ = 0 (where as usual we write the group operation in the torus additively). Finally, if σ is nontrivial mod 2, then G(σ) = U(2) T(r 2) and h is the nontrivial central ∼ × − σ element of order 2 in U(2).

The following statement is a consequence of the naturality of the above constructions.

5.5 Proposition Suppose that G is a connected compact Lie group with maximal torus T , maximal torus normalizer N , Weyl group W , root subgroups N , and markings h as above. Suppose that x N has image w W . { σ} { σ} ∈ ∈ Then for any reﬂection σ W , ∈ −1 xNσx = Nwσw−1 −1 xhσx = hwσw−1 In particular, (T, W, h ) is a marked reﬂection torus. { σ }

5.6 Remark Let L denote π1T and let R be the root system in L associated to the above marked reﬂection torus (see Propositions 2.16 and 2.17). It is not hard to identify the dual root system R# in L# (see Section 2.21) with the root system usually associated to G [31, VIII-8, VI]. The correspondence G R# 7→ produces a bijection between isomorphism classes of connected compact Lie groups and isomorphism classes of root systems (see Section 2.23).

The marking hσ is defined above (Definition 5.3) in terms of Nσ . However, it is − not hard to see that hσ is the only marking for σ in the intersection of T0 (σ) with the square of the nonidentity component of N(σ). Since N(σ) is the cen- + tralizer in N of T0 (σ), the element hσ , and hence the marked reflection torus

eometry & opology, Volume 9 (2005) G T 1358 W G Dwyer and C W Wilkerson

(T, W, h ) and the root system R# , are determined by the group structure { σ } of N . In this way we recover the result from [8] and [30] that G is determined up to isomorphism by N .

The main result of this section is the following one.

5.7 Theorem Let G be a connected compact Lie group, and (T, W, h ) { σ } the marked reﬂection torus derived from G (Proposition 5.5). Then the normalizer of T in G is isomorphic to the normalizer extension ν(T, W, h ) (see { σ } Section 3).

This is a consequence of the following theorem, due essentially to Tits. See Section 4.6 for the notation “prod(n; x,y)”.

5.8 Theorem Suppose that G is a connected compact Lie group with maximal torus T , maximal torus normalizer N , Weyl group W , root subgroups N , { σ} and markings h . Choose a set s of simple reflections in W , and let { σ} { i} m be the corresponding Coxeter matrix. For each i, choose an element x { i,j} i in the nonidentity component of Nsi . Then the elements xi satisfy the following relations: 2 xi = hsi x tx−1 = s (t), t T (5.9) i i i ∈ prod(m ; x ,x ) = prod(m ; x ,x ) for i = j. i,j i j i,j j i 6 Proof of Theorem 5.7 (given Theorem 5.8) Let ν(W ) denote the normalizer extension in question. According to Definition 3.5, Proposition 4.4, and Section 4.6, ν(W ) is generated by the torus T together with symbols qi , one for each simple reflection si , subject to the following relations: 2 qi = hsi q tq−1 = s (t) t T (5.10) i i i ∈ prod(m ; q ,q ) = prod(m ; q ,q ) for i = j. i,j i j i,j j i 6 By Theorem 5.8, there is a homomorphism φ: ν(W ) N which is the identity → on T and sends qi to xi . This is an isomorphism because it is part of a map 1 T ν(W ) W 1 −−−−→ −−−−→ −−−−→ −−−−→ k φ k    1 T N W 1 −−−−→ y −−−−→ y −−−−→ y −−−−→ of exact sequences.

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1359

The proof of Theorem 5.8 depends on two auxiliary results. Suppose that G G is a finite covering of connected compact Lie groups, and that T is a → maximal torus in G. Then the image T of T in G is a maximal torus for G, thee normalizer N of T in G maps onto the normalizer N of T in G, ande the induced map N/Te N/T is an isomorphism.e In particular, G and G have → isomorphic Weyle groups,e whiche we will denote by the same letter W . The lattice correspondinge to Te is a subgroup of finite index in the lattice correspondinge to T , and it follows that an element of W acts as a reflection on T if and only if it acts as a reflectione on T . The following statement is clear. e 5.11 Proposition Suppose that p: G G is a finite covering of connected → compact Lie groups with Weyl group W . Let T be a maximal torus in G and T G the image maximal toruse in G. For each reflection σ W , let ⊂ ∈ Nσ , hσ (respectively, Nσ , hσ ) denote the root subgroupe and marking in Ge (respectively, G) corresponding to σ. Then p(Nσ)= Nσ and p(hσ)= hσ . e e e 5.12 Proposition Suppose that G is a connected,e simply connectede compact Lie group with maximal torus T , torus normalizer N , Weyl group W , root subgroups N , and markings h . Let s , ,s be a set of simple reflections { σ} { σ} { 1 · · · ℓ} for W . Then if i and j are distinct integers between 1 and ℓ, N N = 1 . si ∩ sj { }

Proof The image of N in W is 1,σ , and so if i = j the intersection σ { } 6 N N is contained in T . Since N T is the identity component T −(σ) of si ∩ sj σ ∩ 0 T −(σ), we must show that for i = j , T −(s ) T −(s )= 0 . 6 0 i ∩ 0 j { } Consider the root system R in L = π1T associated to the marked reﬂection torus (T, W, h ) (Remark 5.6). It follows from the fact that G is simply { σ } connected that the roots in R (equivalently, the coroots in R# ) span L; see [31, VIII.1], where Γ(G) denotes what we call L and Hα denotes the set of # { } coroots in R . Let ai be the root corresponding to si . Since all of the roots in R are integral linear combinations of the ai [31, V.8], it is clear that the ai span L. In view of the fact that G is simply connected, the number of simple roots is equal to the rank of G, and so a ,...,a form a basis for L. It { 1 ℓ} follows immediately that L−(s )= a , and so L = L−(s ). Tensoring with i h ii ∼ ⊕i i T(1) gives that T = T −(σ ), which gives the desired result. ∼ ⊕i 0 i

Proof of Theorem 5.8 If the theorem is true for G, it is true for the product G T(n). By Proposition 5.11, if the theorem is true for G and A is a ﬁnite × subgroup in the center of G, it is also true for G/A. Now any connected compact Lie group G can be written as (K T(n))/A, where K is simply ×

eometry & opology, Volume 9 (2005) G T 1360 W G Dwyer and C W Wilkerson connected and A is a ﬁnite subgroup of the center of K T(n). It is enough, × then, to work in the case in which G is simply connected.

In SU(2), the square of any element in the nonidentity component of the torus normalizer is equal to the unique nontrivial central element of the group, so it follows from the choice of h that x2 = h . Since x N projects to s W , si i si i ∈ i ∈ it is also clear that for t T , x tx−1 = s (t). What remains to check is whether ∈ i i i prod(m ; x ,x ) prod(m ; x ,x )−1 = 1 for i = j. (5.13) i,j i j i,j j i 6 (The right hand side of this equation will actually lie in the torus T , but we continue to designate the identity element by 1 because it is the identity of the nonabelian group N .) We follow [34, 3.3]. Let x = prod(m 1; x ,x ), and i,j − j i let k denote j or i, according to whether mi,j is even or odd. The desired relation can be rewritten −1 −1 xxjx xk = 1. (5.14) −1 Let w be the image of x in W . Then wsjw sk = 1 (see Section 4.6), so that xN x−1 N . Since x N and x N it follows that the left hand side sj ⊂ sk j ∈ sj k ∈ sk of (5.14) lies in Nsk . On the other hand, one can also rewrite relation (5.13) as

prod(m ; x ,x ) prod(m ; x ,x )−1 = 1 for i = j. (5.15) i,j j i i,j i j 6 Let l denote i or j , according to whether mi,j is even or odd. The same reasoning as before shows that the left hand side of (5.15) lies in Nsl . The conclusion is that the right hand side of (5.14) lies in N N . Since G is sk ∩ sl simply connected, we are done by Proposition 5.12.

6 Lattices, tori, and extensions at the prime 2

In this section we copy the results of Sections 2 and 3, replacing Z by the ring Z2 of 2–adic integers; however, we do not formulate the notion of a root system over Z2 . The proofs are omitted.

6.1 Marked complete reflection lattices A complete lattice L is a finitely generated free module over the ring Z2 of 2–adic integers. A reflection on L is an automorphism σ of L which is conjugate in Aut(Q L) = GL(b r, Q ) to a ⊗ ∼ 2 diagonal matrix diag( 1, 1,..., 1). b − b b

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1361

6.2 Definition Suppose that σ is a reflection on a complete lattice L. A strict marking for σ is a pair (b, β), where b L and β : L Z is a homomorphism ∈ → 2 such that for any x L, σ(x) = x + β(x)b. Two strict markingsb (b, β) and ∈ (b′, β′) are equivalent if (b, β) = (ub, u−1βb) for a unitb u Z . A marking for σ ∈ 2 is an equivalence class b(b, β) of strict markings. { } 6.3 Remark As in Remark 2.18, a reflection has two markings if it is trivial mod 2, and one marking otherwise.

Suppose that L is a complete lattice, σ is a reflection on L with marking (b, β) , and w is an automorphism of L. Let w (b, β) denote the marking { } · { } for wσw−1 givenb by (w(b), β w−1) . b { ◦ } b 6.4 Definition A complete reflection lattice is a complete lattice L together with a finite subgroup W of Aut(L) which is generated by the reflections it contains. A marked complete reflection lattice is a complete reflection latticeb (L,W ) together with markings (b , β )b, one for each reflection σ in W , such that { σ σ } for each element w W and reflection σ, w (b , β ) = (b −1 , β b−1 ) . ∈ · { σ σ } { wσw wσw } 6.5 Marked discrete reflection tori Let T(˘ n) denote (Z/2∞)n . A discrete torus is a discrete group isomorphic to T(˘ n) for some n. Any discrete torus T˘ gives a complete lattice Hom(T(1)˘ , T˘); conversely, a complete lattice L gives a discrete torus T(1)˘ L. These two constructions are inverse to one another ⊗ up to natural isomorphism, and induce an equivalence between the categoryb of discrete tori and the categoryb of complete lattices.

6.6 Definition An automorphism σ of a discrete torus T˘ is a reflection (respectively, trivial mod 2) if the induced automorphism of Hom(T(1)˘ , T˘) is a reflection (respectively, trivial mod 2). An element x of T˘ is said to be strongly ˘− ˘− σ–negative if x lies in the maximal divisible subgroup T0 (σ) of T (σ).

6.7 Remark An automorphism σ of T˘ as above is trivial mod 2 if and only if it acts as the identity on T = x T 2x = 0 . If T = T(1)˘ L, then 2 { ∈ | } ⊗ x T˘ is strongly σ–negative if and only if x T(1)˘ L−(σ). ∈ ∈ ⊗ b 6.8 Deﬁnition Suppose that σ is a reﬂection on a discreteb torus T˘. A marking for σ is an element h T˘ such that ∈ (1) h is strongly σ–negative (Remark 6.7), (2) 2h = 0, and

eometry & opology, Volume 9 (2005) G T 1362 W G Dwyer and C W Wilkerson

(3) h =0 if σ is nontrivial mod 2. 6

6.9 Definition A discrete reflection torus is a discrete torus T˘ together with a finite subgroup W of Aut(T˘) which is generated by the reflections it contains. A marked discrete reflection torus is a discrete reflection torus (T,W˘ ) together with markings h T˘, one for each reflection σ in W , such that for w W , σ ∈ ∈ hwσw−1 = w(hσ).

6.10 Proposition If L is a complete lattice, there is a natural bijection between marked reflection structures on L and marked reflection structures on the discrete torus T(1)˘ b L. ⊗ b 6.11 The reflection extensionb (Compare with Section 3.1.) Suppose that (T,W˘ ) is a discrete reflection torus, with associated complete reflection lattice

L. Let Σ be the set of reflections in W , and write Σ = i Σi as a union of conjugacy classes of reflections. For each index i choose a reflection τ in Σ , ` i i letb C be the centralizer of τ in W , and let a L be a nonzero element with i i i ∈ τ (a ) = a . It is clear that each element of C takes a to a , and that i i − i i i ± i C is isomorphic to Z/2 C⊥ , where C⊥ is the subgroupb of C consisting of i × i i i elements which fix ai . Consider the extension 1 Z Z C⊥ Z/2 C⊥ 1. (6.12) → → × i → × i → By Shapiro’s lemma, this gives rise to an extension ρi(W ) of W by Z[Σi], and it is easy to see that up to canonical isomorphism this extension of W does not depend on the choice of representative τi for the conjugacy class Σi .

6.13 Definition The reflection extension ρ(W ) of W is the extension by Z[Σ] given by the sum of the extensions ρi(W ), in other words, by the fibre product of the extensions over W .

6.14 The normalizer extension Let (T,W˘ ) be a discrete reflection torus, and Σ the set of reflections in W . Given a marking h for (T,W˘ ), ie a { σ} marked reflection torus = (T˘ , W, h ), there is a W –map f T : Z[Σ] T˘ T { σ} → which sends a reflection σ to hσ .

6.15 Definition Suppose that = (T˘ , W, h ) is a marked discrete reflec- T { σ } tion torus. The normalizer extension ν˘(T, W, h ) of W by T˘ is the image { σ } under f T : Z[Σ] T˘ of the reflection extension ρ(W ). →

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1363

7 Classiﬁcation of marked complete reﬂection lattices

In this section we prove Proposition 7.4, which roughly states that, one example aside, every marked complete reﬂection lattice arises from a marked reﬂection lattice, ie from a classical root system.

It will be convenient to consider pairs (A, W ) in which A is an abelian group and W is a group acting on A; one example is a reflection lattice (L,W ). It is clear what it means for two such objects to be isomorphic. The product (A, W ) (A′,W ′) of two is the pair (A A′,W W ′), with W W ′ acting × × × × on A A′ in the obvious way. Note that the product of two complete and/or × marked reflection lattices is naturally a reflection lattice of the same type. If (L,W ) is a marked reflection lattice, then (Z L,W ) is naturally a marked 2 ⊗ complete reflection lattice.

7.1 Definition A marked complete reflection lattice is of Coxeter type if it is isomorphic to (Z L,W ) for some marked reflection lattice (L,W ). 2 ⊗

Let (L∆,W∆) be the complete reﬂection lattice of rank 3 associated to the exceptional 2–compact group DI(4) [10, 24].

7.2 Remark The group W is isomorphic to Z/2 GL(3, F ); the central ∆ × 2 Z/2 in W acts by 1 on L , and W acts on Z/2 L = (Z/2)3 via the ∆ − ∆ ∆ ⊗ ∆ ∼ natural representation of GL(3, F2). These last two facts essentially determine the action of W∆ on L∆ [10, Section 4]. Every reﬂection in W∆ is nontrivial mod 2, and so has a unique marking (Remark 6.3); in particular, there is a unique way to give (L∆,W∆) the structure of marked complete reﬂection lattice.

7.3 Deﬁnition A marked complete reﬂection lattice is of type DI(4) if it is isomorphic to a product of copies of (L∆,W∆).

7.4 Proposition Every marked complete reﬂection lattice (L,W ) is isomorphic to a product (L ,W ) (L ,W ), in which the ﬁrst factor is of Coxeter 1 1 × 2 2 type and the second is of type DI(4). b b b This depends on a few lemmas.

eometry & opology, Volume 9 (2005) G T 1364 W G Dwyer and C W Wilkerson

7.5 Definition A marked complete reflection lattice (L,W ) is rationally of Coxeter type if there is a reflection lattice (L,W ) such that (Q L,W ) is ⊗ isomorphic as a reflection lattice to (Q L,W ). b 2 ⊗ b 7.6 Lemma Suppose that (L,W ) is a marked complete reflection lattice which is rationally of Coxeter type. Then (L,W ) is of Coxeter type. b Proof Since (L,W ) is rationally ofb Coxeter type, there is a lattice L′ and an action of W on L′ such that (Q L,W ) is isomorphic to (Q L′,W ); we ⊗ 2 ⊗ use this isomorphismb to identify Q L with Q L′ . Under this isomorphism, ⊗ 2 ⊗ L′ = Z L′ is a complete lattice inbQ L preserved by W . Adjust L′ by a 2 ⊗ ⊗ power of 2 so that L L′ . Now defineb L by the pullback diagram: ⊂ b b L L b b −−−−→ b ′ ′ L L y −−−−→ y The group L′/L is a finite 2–group, and the map L′ L′ L′/L is surjective, → → this last by Nakayama’s lemma and the factb that Z/2 L′ = Z/2 L′ . It ⊗ ∼ ⊗ follows thatbL′b/L is isomorphic to L′/L, and that L isb a latticeb inb Q L′ with ⊗ Z L = L. In particular, (L,W ) is isomorphic to (Z L,W ). Theb fact 2 ⊗ ∼ 2 ⊗ that under this isomorphism the markingb b on L is derived from a marking on L follows fromb the observationb that possible markings of (L,W ) correspond bijectively to markings of (L,W ). b b 7.7 Lemma Suppose that (L,W ) is a complete reflection lattice, such that (Q L,W ) decomposes as a product (V ,W ) (Q L ,W ). Then (L,W ) ⊗ 1 1 × ⊗ ∆ ∆ decomposes as the product (bL V ,W ) (L ,W ) of complete reflection ∩ 1 1 × ∆ ∆ lattices.b If (L,W ) is marked, then there is a unique way to mark theb two factors of the product in suchb a way that the product decomposition respects the markings.b

Proof Write Q L = V V and W = W W , where V = Q L and ⊗ ∼ 1 × 2 ∼ 1 × 2 2 ∼ ⊗ ∆ W = W . Let L = L V , so that L is a complete lattice which is a module 2 ∼ ∆ i ∩ i i over W and Q Lb = V . By elementary rank considerations, if σ W is i ⊗ i ∼ i ∈ a reflection, thenb eitherb σ W or bσ W . Consequently, W and W are ∈ 1 ∈ 2 1 2 generated by the reflectionsb they contain, and both (L1,W1) and (L2,W2) are complete reflection lattices. Now adjust the copy of L∆ in V2 by multiplication by a power of 2 so that L L but L is not containedb in 2L . Theb induced ∆ ⊂ 2 ∆ 2 map f : Z/2 L Z/2 L is then a nontrivial map between W –modules ⊗ ∆ → ⊗ 2 2 b b b eometry & opology, Volume 9 (2005) G T Normalizers of tori 1365 which are F –vector spaces of the same dimension; since Z/2 L is simple 2 ⊗ ∆ as a W2 –module (Remark 7.2), the map f is an isomorphism and it follows from Nakayama’s lemma that L2 is isomorphic to L∆ as a W2 –module. Let L = Z/2 L and L = Z/2 L . Consider the two W –module maps ⊗ i ⊗ i b g : L L L g′ : L L L. b 1 × 2b→ 1 × 2 → Again by Nakayama’s lemma,b b in orderb to prove that g is an isomorphism it is enough to prove that g′ is an isomorphism, or even, by a rank calculation, that g′ is injective. Let g′ denote the natural map L L and let K be the i i → kernel of g′ . It is clear from the definitions that g′ is injective (ie, if x L and i ∈ 2x L then x L ), so that the two maps K L and K L are injective. ∈ i ∈ i → 1 → 2 Since W2 acts trivially on L1 , W2 must act trivially on K , and so the factb that K = b0 follows fromb the fact that L = Z/2 L has no submodules with a { } 2 ∼ ⊗ ∆ trivial W2 –action. The quickest way to obtain the statement about markings is to interpret the markings as elements of the associated discrete tori (6.10).

7.8 Proposition The only irreducible finite reflection groups over Q2 are the classical Weyl groups, with their standard reflection representations, and the group W∆ , with the reflection representation derived from DI(4).

Proof This follows from [7]. Clark and Ewing observe that the Schur index of a reflection representation is 1, so that the representation is defined over its character field, that is, the field extension of Q generated by the characters of the group elements. They then determine the character fields of all of the irreducible complex reflection groups. The only character fields contained in Q2 are the fields derived from classical Weyl group representations (which are all Q) and the field derived from the reflection representation of W∆ (which is Q(√ 7)). The complex reflection representation of W is unique up to − ∆ conjugacy, which easily implies that the associated Q2 –reflection representation is also unique.

Proof of Proposition 7.4 Let (L,W ) be a complete reflection lattice. Ac- cording to the classification of 2–adic rational reflection groups (see Proposi- tion 7.8), we can write (Q L,Wb) as a product (V ,W ) (Q L ,W )k ⊗ 1 1 × ⊗ ∆ ∆ for some k 0, where W is a classical Weyl group. Repeated applications ≥ 1 of Lemma 7.7 give that (L,Wb) is equivalent as a marked complete reflection lattice to (L ,W ) (L ,W )k , where (L ,W ) is rationally of Coxeter type. 1 1 × ∆ ∆ 1 1 The result follows from Lemmab 7.6. b b

eometry & opology, Volume 9 (2005) G T 1366 W G Dwyer and C W Wilkerson

8 Completions of compact Lie groups

In this section we give some results about completions of Lie groups that are used in Section 9. Suppose that G is a connected compact Lie group of rank r and that X is the 2–compact group obtained by taking the 2–completion of G (so that BX is the 2–completion of BG). Let (T, W, h ) be the marked { σ } reflection torus obtained from G, and T˘ the 2–primary torsion subgroup of T (1.11), so that T˘ is a discrete torus of rank r. It is easy to see that (T˘ , W, h ) { σ } is a marked discrete reflection torus. Let ν(W ) be the normalizer extension of W by T determined by (T, W, h ) (Definition 3.5) andν ˘(W ) the parallel { σ } normalizer extension of W by T˘ (Definition 6.15). Essentially by construction there is a homomorphismν ˘(W ) ν(W ) which lies in a map of exact sequences: → 1 T˘ ν˘(W ) W 1 −−−−→ −−−−→ −−−−→ −−−−→ ⊂ k    1 T ν(W ) W 1 −−−−→ y −−−−→ y −−−−→ y −−−−→ By Theorem 5.7, the inclusion T G extends to a homomorphism ν(W ) G. → → Combining these homomorphisms with the completion map BG BX gives → composite maps BT˘ BT BG BX and Bν˘(W ) Bν(W ) BG → → → → → → BX , which, in the language of [11], correspond to homomorphisms T˘ X and → ν˘(W ) X . →

8.1 Lemma In the above situation, T˘ X is a maximal discrete torus → for X , ν˘(W ) is the normalizer of T˘ in X , and ν˘(W ) X is the natural → homomorphism.

See Aside 9.1 for a description of the maximal discrete torus in a 2–compact group.

Proof of Lemma 8.1 Consider ﬁrst the chain of maps

BT˘ BT BG BX. → → → The left hand map induces an isomorphism on mod 2 homology (see [11, Sec- tion 6]), as does the completion map BG BX . Let X/T˘ denote the homo- → topy fibre of BT˘ BX , G/T the homotopy fibre of BG BX , and G/T˘ → → the homotopy fibre of BT˘ BG. Let T denote the closure [12, Section 3] of → T˘ (ie BT is the 2–completion of BT˘), so that BT˘ BX extends canonically → b b eometry & opology, Volume 9 (2005) G T Normalizers of tori 1367 to BT BX , and denote the homotopy fibre of this last map by X/T . It → follows that all of the maps in the diagram b X/T X/T˘ G/T˘ G/T b ← ← → induce isomorphisms on mod 2 homology. In particular, X/T has finite mod 2 homology, and so T X isb a monomorphism [11, 3.2, 9.11]. It is straight- → forward to argue that the Euler characteristic χQ2 (X/T ) [11,b 1.5] is the same as the rational Eulerb characteristic χQ(G/T ); since this last is nonzero [14, 5.3], the map T X is a maximal torus for X [12,b 2.15]. Since T˘ T is → → a discrete approximation, T˘ X is a 2–discrete maximal torus for X . The → statements aboutb ν ˘(W ) amount to an identification of W with the Weyl groupb W of X . There is certainly a map W W , since W maps to the self- X → X equivalences of BT over BG and hence by completion to the self-equivalences of BT over BX (see [11, 9.5]). The map is a monomorphism because the completion map GL(r, Z) = Aut(π T ) Aut(π T ) = GL(r, Z ) is injective, and ∼ 1 → 1 ∼ 2 consequentlyb an isomorphism by counting: #(W ) = χQ(G/T ) [14, 5.3], while

#(WX )= χQ2 (X/T ) [11, 9.5]. b

We will also need tob refer to the following results. 8.2 Proposition Suppose that G is a connected compact Lie group, that X is the 2–completion of G, and that π is a ﬁnite 2–group. Then the map Bf which sends a homomorphism f : π G to the composite Bπ BG BX → −−→ → induces a bijection between conjugacy classes of homomorphisms π G and → homotopy classes of maps Bπ BX . → Proof This amounts to combining [9, 1.1] with the fact that the completion map BG BX induces a bijection from homotopy classes of maps Bπ BG → → to homotopy classes of maps Bπ BX . This last follows from [9, 2.4]. → 8.3 Proposition Suppose that G is a connected compact Lie group with maximal torus T , and that S˘ T is a discrete torus with topological closure ⊂ ˘ S. Let X be the 2–completion of G, GS˘ the centralizer of S in G, GS the ˘ centralizer of S in G, and XS˘ the centralizer of S in X . Then the natural map G G is an isomorphism, while the natural map BG BX is a S → S˘ S˘ → S˘ 2–completion map.

Proof It is elementary that G G is an isomorphism. The fact that S → S˘ BG BX is a 2–completion map follows from the main theorem of [25] S˘ → S˘ and the argument in the proof of [9, 2.4]. Note that XS˘ is 2–complete by [11, 5.7, 6.21].

eometry & opology, Volume 9 (2005) G T 1368 W G Dwyer and C W Wilkerson

9 2–compact groups

In this section we associate a marked discrete reﬂection torus to a connected 2– compact group X (Proposition 9.10), and show that the associated normalizer extension of the Weyl group is isomorphic to the normalizer of a maximal discrete torus in X (Proposition 9.12).

9.1 Aside A complete torus T is a 2–compact group with the property that its classifying space BT is equivalent to the 2–completion of the classifying space of an ordinary torus. Anyb 2–compact group X has a maximal (complete) torus, which is technicallyb a complete torus T together with a monomorphism T X with the property that the Euler characteristic of X/T is nonzero [11, → Section 8]. For such a maximal torus, the spaceb of self-maps of BT over W BXb is homotopically discrete, and its monoid of componentsb is a ﬁnite group W called the Weyl group of X . The monoid acts on BT , and theb Borel W construction of this action is denoted BN(T ); this lies in a ﬁbration sequence b BT BN(T ) BW. → →b The loop space of BN(T ) is called the normalizer of T in X . b b Any complete torus has a discrete approximation, which is a discrete torus T˘ b together with a homomorphism T˘ T with the property that the induced → map BT˘ BT induces an isomorphism on mod 2 homology. If T X is → → a maximal torus for X , then the compositeb T˘ T X is called a maximal → → discrete torus forb X . See [12, 3.13] for an explanation of how in thisb case the approximation T˘ T can be extended to give a discreteb approximation N˘ for → N(T ): N˘ is a discrete group which lies in a short exact sequence b 1 T˘ N˘ W 1 , b { }→ → → → { } and it is called the normalizer of T˘ in X .

Suppose that X is a connected 2–compact group of rank r with maximal discrete torus T˘, Weyl group W , and discrete torus normalizer N˘ . Then ˘ ˘ ˘ T ∼= T(r) and the conjugation action of W on T identifies W with a subgroup of Aut(T˘) generated by reflections [11, 9.7]. In particular, (T,W˘ ) is a 2– discrete reflection torus. For each reflection σ W we will construct a subgroup N˘ of N˘ , called the ∈ σ root subgroup associated to σ. We will also single out a specific element h T˘, σ ∈ the marking associated to σ. The technique is to mimic in homotopy theory the constructions of Section 5.

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1369

˘+ ˘ Given σ, let T0 (σ) denote the maximal divisible subgroup of T on which σ acts trivially. It is easy to see that T˘+(σ) is a discrete torus of rank r 1; 0 − if L is the complete lattice corresponding to T˘, then under the isomorphism ˘ ˘ ˘+ ˘ + T ∼= T(1) L, T0 (σ) is given by T(1) L (σ). Let X(σ) denote the centralizer ˘+ ⊗ ⊗ of bT0 (σ) in X . b b 9.2 Aside The inclusion homomorphism i: T˘+(σ) T˘ X is by deﬁnition 0 → → represented by a map Bi: BT˘+(σ) BX of classifying spaces. Note that the 0 → domain of Bi is a space of type K(π, 1). The centralizer X(σ) is by deﬁnition the 2–compact group whose classifying space BX(σ) is the mapping space ˘+ component Map(BT0 (σ), BX)Bi .

Then X(σ) is a connected 2–compact subgroup of X [12, 7.8] with maximal discrete torus T˘, and Weyl group given by the subgroup 1,σ of W [12, 7.6]. { } The normalizer of T˘ in X(σ) can be identiﬁed with the algebraic centralizer of ˘+ ˘ ˘ ˘ T0 (σ) in N ; we denote this group by N(σ). Note that N(σ) is an extension of Z/2 by a discrete torus, and so is a 2–discrete toral group [11, 6.5].

˘+ 9.3 Aside Let A = BT0 (σ). The classifying space maps A = BT˘+(σ) u BT˘ v BN˘ w BX 0 −→ −→ −→ induce maps Map(A, BT˘) Map(A, BN˘ ) Map(A, BX) , (9.4) u → vu → wvu The mapping space component on the far right is BX(σ). By covering space theory, Map(A, BT˘)u is canonically equivalent to BT˘, the equivalence being given by evaluation at the basepoint of A. Similarly, Map(A, BN˘ )vu is the ˘+ ˘ classifying space of the algebraic centralizer of T0 (σ) in N . It follows from [12, 7.6] that the composite map in (9.4) represents the inclusion in X(σ) of a maximal discrete torus, and that the right-hand map represents the inclusion of the normalizer of this torus.

The only simply-connected 2–compact group of rank 1 is the two-completion SU(2) of SU(2) (see [16]); the center of SU(2) is Z/2 and the adjoint form is the 2–completion SO(3) of SO(3). Given this, it follows from the product splitting theoremb for p–compact groups [13] thatb X(σ) is given by one of the following three possibilitiesb (see [12, 7.7], [21]), where T(k) denotes a 2–complete torus of rank k and U(2) is the 2–completion of U(2): b (1) A product SU(2) T(r 1) b × − b b eometry & opology, Volume 9 (2005) G T 1370 W G Dwyer and C W Wilkerson

(2) A product SO(3) T(r 1) × − (3) A product U(2) T(r 2) b × b − Recall that T˘−(σ) denotes the maximal divisible subgroup of T˘−(σ); this group 0 b b is a discrete torus of rank 1.

9.5 Definition The root subgroup N˘σ is given by N˘ = x N˘ (σ) x is conjugate in X(σ) to some y T˘−(σ) . σ { ∈ | ∈ 0 } 9.6 Aside This definition is couched in the language of [11], but what it means concretely is the following. Suppose that x N˘ (σ) has order 2n ; note that by ∈ [11, 6.19], N˘(σ) is a union of finite 2–groups. Represent x by χ(x): Z/2n → N˘ (σ). Then x N˘ if and only if the composite ∈ σ Bχ(x) BZ/2n BN˘ (σ) BX(σ) −−−−→ → lifts up to homotopy to a map BZ/2n BT˘−(σ). → 0 Given that X(σ) is the 2–completion of a compact Lie group, the following statement comes from combining Lemma 5.2 with the discussion in Section 8. Lemma 9.9 is proved similarly.

9.7 Lemma The set N˘σ is a subgroup of N˘ (σ) isomorphic to an extension of Z/2 by a rank 1 discrete torus. More precisely, N˘ T˘ = T˘−(σ), and the natural σ∩ 0 map N˘ /T˘−(σ) N˘ (σ)/T˘ = Z/2 is an isomorphism. If x,y N˘ T˘−(σ), σ 0 → ∼ ∈ σ \ 0 then x2 = y2 T˘−(σ). ∈ 0 9.8 Definition The marking h T˘−(σ) is defined to be the image of N˘ σ ∈ 0 σ \ T˘−(σ) under the squaring map x x2 . 0 7→ 9.9 Lemma The element h T˘ is a marking for σ in the sense of Defini- σ ∈ tion 6.8.

9.10 Proposition Suppose that X is a connected 2–compact group with maximal discrete torus T˘, discrete torus normalizer N˘ , Weyl group W , root subgroups N˘ , and markings h as above. Suppose that x N˘ has image { σ} { σ} ∈ w W . Then for any reﬂection σ W , ∈ ∈ ˘ −1 ˘ xNσx = Nwσw−1 −1 xhσx = hwσw−1 In particular, (T˘ , W, h ) is a marked discrete reﬂection torus. { σ }

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1371

Proof As with Proposition 5.5, the idea here is that conjugation with x gives a symmetry of the whole situation, but there is something to check. Let c denote conjugation with x, and let τ = wσw−1 . All of the arrows in the following commutative diagram are clear, except for the right hand vertical one:

BT +(σ) BN˘ X 0 −−−−→ −−−−→ Bc Bc Bc′ + ˘  BT0(τ) BN X y −−−−→ y −−−−→ y If the right hand arrow exists, then taking mapping spaces (Aside 9.2) gives a commutative diagram BN˘(σ) BX(σ) −−−−→ Bc   BN˘(τ) BX(τ) y −−−−→ y which, in view of Definitions 9.5 and 9.8, yields the desired result. The issue then is to extend the self equivalence Bc of BN˘ to a self equivalence Bc′ of BX . One way to achieve this is to represent the map N˘ X by a map → U V of simplicial groups; the map c can then be realized by conjugation with → a vertex of U , and it extends to an automorphism of V given by conjugation with the image vertex. Another approach is to treat c as part of the conjugation action of N˘ on itself. Taking classifying spaces translates this into a basepoint- preserving action of N˘ on BN˘ , an action which is homotopically captured by the associated sectioned fibration over BN˘ with fibre BN˘ . It is easy to see that this fibration is just the product projection BN˘ BN˘ BN˘ , with section × → given by the diagonal map. This fibration evidently extends to the product fibration BX BX BX . × →

9.11 Remark It is natural to conjecture that the above marked discrete re- ﬂection torus, or equivalently the associated marked complete reﬂection lattice (Proposition 6.10), determines X up to equivalence.

The main result of this section is the following one.

9.12 Proposition Suppose that X is a connected 2–compact group, and that (T˘ , W, h ) is the marked discrete reﬂection torus associated to X (see { σ} Proposition 9.10). Then the normalizer N˘ of T˘ in X is isomorphic to the normalizer extension ν˘(T˘ , W, h ) (Deﬁnition 6.15). { σ }

eometry & opology, Volume 9 (2005) G T 1372 W G Dwyer and C W Wilkerson

The proof breaks up into two cases. A connected 2–compact group X is said to be of Coxeter type (respectively, of type DI(4)) if the associated complete marked reflection lattice (Propositions 9.10 and 6.10) is of Coxeter type (respectively, of type DI(4)); see Definitions 7.1 and 7.3. Note that this is a property of the underlying complete reflection lattices, or equivalently of the action of W on T˘; the markings come along automatically (see Remark 7.2 and the proof of Lemma 7.6).

9.13 Proposition Any connected 2–compact group X can be written as a product X X of 2–compact groups, where X is of Coxeter type and X 1 × 2 1 2 is of type DI(4).

Proof This is a consequence of Proposition 7.4 and [13, 6.1].

A connected 2–compact group X is said to be semisimple if its center is finite (equivalently, if π1X is finite). If X is any connected 2–compact group, the universal cover X is semisimple, and it follows from [13] and [21] that X is obtained from the product of X and a 2–complete torus by dividing out [11, 8.3] by a finite centrale subgroup. We express this by saying that X is a central product of X and a torus. Thee semisimple rank of X is the rank of X ; or equivalently the rank of X minus the rank of the center of X . There are analogous notionse for compact Lie groups. e The rank 1 case of the following statement follows from [16]; the rank 2 case is proved by Bauer et al cite[6.1]rKN.

9.14 Proposition If X is a semisimple 2–compact group of rank 2, then ≤ X is the 2–completion of a connected compact Lie group.

If G is a semisimple Lie group, the center of G is a ﬁnite abelian group (G). Z It is easy to see that the center of the corresponding 2–compact group G is the quotient of (G) by the subgroup of elements of odd order; for instance, com- Z bine the parallel calculations [12, 7.5] and [14, 8.2] with Lemma 8.1. Thisb implies that any central product of G with a 2–complete torus is the 2–completion of some central product of G with a torus. In conjunction with Proposition 9.14, this observation gives the following.b

9.15 Proposition If X is a connected 2–compact group of semisimple rank at most 2, then X is the 2–completion of a connected compact Lie group.

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1373

Suppose that X is a connected 2–compact group with marked discrete reflection torus (T˘ , W, h ) and discrete torus normalizer N˘ as above. Let A be a { σ } subgroup of T˘, XA the centralizer of A in X , and WA the subgroup of W consisting of all elements which leave A pointwise fixed. Then T˘ is a maximal discrete torus for XA , and WA is the Weyl group of XA . The normalizer N˘A of T˘ in XA is the centralizer of A in N˘ , or equivalently the pullback over W W of the surjection N˘ W . We assume that X is connected; A ⊂ → A this is always true if A is a discrete torus, and the general criterion is in [12, 7.6]. Under this assumption, WA is generated by the reflections it contains [12, 7.6], and so (T,W˘ ) is a discrete reflection torus. Let h denote A { σ}σ∈WA the set of markings for the reflections in W obtained from the given set h A { σ} of markings for the reflections in W . It is clear that (T,W˘ , h ) is A { σ}σ∈WA a marked reflection torus. For each reflection σ W , let N˘ be the root ∈ A σ subgroup for σ computed in X , and N˘ N˘ the root subgroup for σ A,σ ⊂ A computed in X . Given the inclusion N˘ N˘ , both of these root subgroups A A ⊂ can be considered as subgroups of N˘ .

9.16 Lemma In the situation described above, (T,W˘ , h ) is the dis- A { σ}σ∈WA crete marked reﬂection torus associated to XA by Proposition 9.10. Moreover, for each reﬂection σ W , N˘ = N˘ as a subgroup of N˘ . ∈ A A,σ σ

Proof By Definition 9.8, the first statement follows from the second. Let B = ˘+ T0 (σ), X(σ) the centralizer of B in X , and XA(σ) the centralizer of B in XA . Both X(σ) and XA(σ) are connected [12, 7.8], and the monomorphism [11, 5.2, 6.1] X X induces a monomorphism X (σ) X(σ). The monomorphism A → A → is an equivalence by [12, 4.7], since both XA(σ) and X(σ) have Weyl group Z/2 generated by σ. A similar argument shows that the centralizer N˘A(σ) of B in N˘A is the same as the centralizer N˘ (σ) of B in N˘ : both groups have T˘ as a maximal normal divisible abelian subgroup, and the quotient of each group by T˘ is 1,σ . The second statement of the lemma thus follows from { } Definition 9.5

9.17 Lemma Suppose that G is a connected compact Lie group, and X the 2–compact group obtained as the 2–completion of G. Let W be the Weyl group of G (or of X ), N the normalizer of a maximal torus in G, N˘ the normalizer of a maximal discrete torus T˘ in X , and Nσ (respectively, N˘σ ) the root subgroup of N (respectively, N˘ ) corresponding to a reﬂection σ W . ∈ Then under the identiﬁcation given by Lemma 8.1 of T˘ as a subgroup of T and N˘ as a subgroup of N , N˘ = N˘ N . σ ∩ σ

eometry & opology, Volume 9 (2005) G T 1374 W G Dwyer and C W Wilkerson

+ ˘+ ˘ Proof Let T0 (σ) T (respectively, T0 (σ) T ) be as in Section 5 (respec- ⊂ ˘+ ⊂ tively, as above). It is easy to see that T0 (σ) is the group of 2–primary torsion + + ˘+ elements in T0 (σ), so that T0 (σ) is the topological closure of T0 (σ) in T , ˘+ ˘ + ˘ and T0 (σ) = T T0 (σ). Let N(σ) (respectively, N(σ)) be the centralizer + ∩ ˘+ ˘ of T0 (σ) in N (respectively, the centralizer of T0 (σ) in N ). It follows that ˘ ˘ + N(σ)= N N(σ). Let G(σ) (respectively, X(σ)) be the centralizer of T0 (σ) ∩ ˘+ in G (respectively, the centralizer of T0 (σ) in X ). There is a commutative diagram: BT˘−(σ) BT −(σ) 0 −−−−→ 0

  BN˘(σ) BN(σ) BG(σ) BX(σ) y −−−−→ y −−−−→ −−−−→ ˘− − ˘ Again, T0 (σ) is the 2–primary torsion subgroup of T (σ). Note that N(σ) is a union of finite 2–groups (Aside 9.6). According to Proposition 8.3 and Proposition 8.2, an element x N˘(σ) is homotopically conjugate (Aside 9.6) ˘− ∈ in X(σ) to an element of T0 (σ) if and only if x is algebraically conjugate in ˘− G(σ) to an element of T0 (σ). Since x is of order a power of 2, this last occurs − if and only if x is conjugate in G(σ) to an element of T0 (σ). The lemma follows from the definitions of the root subgroups Nσ (Definition 5.1) and N˘σ (Definition 9.5).

Proof of Proposition 9.12, Coxeter case Let (L,W ) be the marked complete reflection lattice associated to X (Proposition 6.10), and (L,W ) a marked reflection lattice such that (L,W ) = (Z L,W ).b Associated to (L,W ) is a ∼ 2 ⊗ marked reflection torus (T, W, h′ ). The group T˘ = L T(1)˘ is naturally { σ } ∼ ⊗ isomorphic to the 2–primaryb torsion subgroup of T = L T(1), and under ∼ ⊗ this identification, h′ = h for each reflection σ W .b Consequently, as in σ σ ∈ Section 8, the normalizer extensionν ˘(W ) =ν ˘(T˘ , W, h ) is a subgroup of the { σ} extension ν(W )= ν(T, W, h ) (Definition 3.5). Choose a set s of simple { σ } { i} reflections in W (see Remark 2.14). By Section 4.6 and Definition 6.15,ν ˘(W ) is generated by T˘, together with symbols qi , one for each simple reflection si , subject to the relations (5.10) (with T replaced by T˘). As in the proof of Theorem 5.7, in order to show thatν ˘(W ) is isomorphic to N˘ , it is enough to find elements x N˘ , one for each simple reflection, such that x projects to i ∈ i si W and such that the xi satisfy relations (5.9) (with T replaced by T˘). For ∈ ˘ each simple reflection si , let Nsi be the corresponding root subgroup, and xi some element in N˘ T −(s ). We claim the elements x satisfy the necessary si \ 0 i i relations. The first relation in (5.9) is clear (Definition 9.8), and the second follows from the fact that x by construction projects to s W . The final i i ∈

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1375 relation involves pairs x ,x of the generating elements. Choose such a pair, { i j} let A be the rank r 2 discrete torus which is the maximal divisible subgroup − of T˘ on which si and sj act as the identity, and let XA be the centralizer of A in X . Then XA is connected [12, 9.8] and of semisimple rank 2. In the ˘ notation of Lemma 9.16, xi (respectively, xj ) is in the root subgroup NA,si ˘ ˘ (respectively, NA,sj ) of N determined by XA , and so in order to verify the final relation between xi and xj we can replace X by XA , or in particular, assume that X has semisimple rank 2. In this case X is the 2–completion of a connected compact Lie group (Proposition 9.15; note that this is a crucial step in the proof, which is due to Kitchloo and Notbohm), and the result essentially follows from Lemma 9.17 and the fact that the relation is satisfied in the Lie group case (Theorem 5.8). The only remaining issue is to show that s ,s is { i j} a set of simple reflections for the Weyl group WA of XA (note that WA is the set of elements in W which pointwise fix A). Let V be the codimension 2 subspace of R L which is pointwise fixed by s ⊗ i and sj ; it is easy to see that WA is the subgroup of W consisting of element which fix V pointwise. The group WA is generated by the reflections it contains (see [19, Theorem 1.2(c)], or note that WA is the Weyl group of a connected 2–compact group). The desired result now follows from [19, Theorem 1.10(a)], given that, in the language of [19], a reflection in W fixes V pointwise if and only if the associated root is orthogonal to V , that is, it lies in the R–span of the roots corresponding to si and sj .

Proof of Proposition 9.12, DI(4) case Suppose that X is of type DI(4). Let x be a nontrivial element of order 2 in T˘, and A the subgroup of T˘ generated by x. Note that the Weyl group W∆ of X acts transitively on the set of such elements x (Remark 7.2), so that the conjugacy class of A as a subgroup of X does not depend on the choice of x. Let XA be the centralizer of A in X , and WA the group of elements in W∆ which fix x. The 2–compact group XA is connected; one quick way to check this is to observe that all of the reflections in W∆ are nontrivial mod 2 (Remark 7.2), and so by [12, Section 7] XA is connected if and only if WA is generated by reflections. But WA is generated by reflections; this can be verified directly, or by observing that in the particular case X = DI(4) [10], the centralizer of such an A is clearly connected, since it is the 2–completion of Spin(7).

By Remark 7.2 and [12, Section 7], the group WA , which is the Weyl group of X , is isomorphic to 1 P (1, 2), where P (1, 2) is the subgroup of GL(3, F ) A {± }× 2 given by matrices which agree with the identity in the ﬁrst column. For cardi- nality reasons WA cannot contain W∆ as a factor, and so by Proposition 7.4

eometry & opology, Volume 9 (2005) G T 1376 W G Dwyer and C W Wilkerson

XA is of Coxeter type. Since the index of WA in W∆ is 7, which is odd, the fact that Proposition 9.12 is true for X follows from Remark 10.2 and the fact proved above that Proposition 9.12 is true for XA .

10 A centralizer lemma

In this section we verify that the normalizer extension construction given in Deﬁnition 6.15 behaves well in a certain sense when it comes to taking central- izers of subgroups of the torus T˘. Our proof of Proposition 9.12 in the DI(4) case depends on this behavior.

We put ourselves in the context of Lemma 9.16; in particular, X is a connected 2–compact group with marked reflection torus (T˘ , W, h ) and discrete torus { σ } normalizer N˘ , A is a subgroup of T˘, XA is the centralizer of A in X and WA is the subgroup of W consisting of elements which leave A pointwise fixed [12, 7.6]. Recall that we are assuming that XA is connected. According to Lemma 9.16, (T,W˘ , h ) is the marked reflection torus associated to X . Let ν(W ) A { σ}σ∈WA A (respectively, ν(WA)) denote the normalizer extension of W (respectively, WA ) obtained from (T˘ , W, h ) (respectively, (T,W˘ , h ). { σ } A { σ}σ∈WA

10.1 Lemma In the above situation, there is a commutative diagram:

1 T˘ ν(W ) W 1 −−−−→ −−−−→ A −−−−→ A −−−−→ k ⊂    1 T˘ ν(W ) W 1 −−−−→ y −−−−→ y −−−−→ y −−−−→ In particular, the extension ν(W ) of W is the pullback over W W of the A A A ⊂ extension ν(W ) of W .

10.2 Remark Let N˘ (respectively, N˘A ) denote the normalizer of T˘ in X (respectively, XA ). It is clear from [12, Section 7] that the extension N˘A of WA is the pullback over WA W of the extension N˘ of W . We can ⊂ ˘ ˘ therefore conclude from Lemma 10.1 that if N ∼= ν(W ), then NA ∼= ν(WA). More interestingly, suppose that the index of WA in W is odd, so that (by a transfer argument) the restriction map H2(W ; T˘) H2(W ; T˘) is injective and → A extensions of W by T˘ are detected on WA . In this case we can conclude that ˘ ˘ if NA ∼= ν(WA), then N ∼= ν(W ).

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1377

10.3 A double coset formula The proof of Lemma 10.1 depends upon a double coset formula for Shapiro companions. If u: H G is an inclusion of → ﬁnite groups and k H2(H; Z), write u (k) for the element of H2(G; Z[G/H]) ∈ # that corresponds to k under Shapiro’s lemma. If v : K G is another sub- → group inclusion, the problem is to compute v∗u (k) H2(K; Z[G/H]). Let x # ∈ α run through a set of (K,H) double coset representatives in G, so that G can be written as a disjoint union

G = KxαH. α a For each index α, let K = K x Hx−1 , and note that there is a natural α ∩ α α isomorphism of K –modules Z[G/H] = Z[K/K ]. (10.4) ∼ ⊕α α Write vα for the map − x 1(–)x v : K x Hx−1 α α H α α → α α −−−−−−→ and uα for the inclusion K K . Then for each index α, uα v∗ (k) α → # α ∈ H2(K; Z[K/K ]), so that, under isomorphism (10.4), the element uα v∗ (k) α ⊕α # α lies in H2(K; Z[G/H]).

10.5 Lemma In the above situation, for any k H2(H; Z), ∈ v∗u (k)= uα v∗ (k). # ⊕α # α Sketch of proof Consider the homotopy ﬁbre square

Bvα α BKα BH −−−−→` `Buα Bu `  Bv  BK BG y −−−−→ y in which the vertical maps are covering spaces with fibre G/H . An element k H2(H; Z) corresponds to an extension of H by Z and thus to a fibra- ∈ 1 tion p: BE BH with fibre BZ = S . The Shapiro companion u#(k) 2 → ∈ H (BG; Z[G/H]) similarly corresponds to some fibration (Bu)#(BE) over BG. One checks that (Bu) (BE) is the fibration whose fibre over x BG is the # ∈ product, taken over y (Bu)−1(x), of p−1(y). In other words, u is a kind of ∈ # geometric multiplicative transfer. By inspection, pulling BE BH back over α → Bvα and applying the multiplicative transfer ( Bu )# gives the same result as pulling (Bu) (BE) back over Bv. The lemma is the algebraic expression ` # ` of this.

eometry & opology, Volume 9 (2005) G T 1378 W G Dwyer and C W Wilkerson

Proof of Lemma 10.1 Let Σ be the set of reflections in W and ΣA the set of reflections in WA , so that Z[Σ] is isomorphic as a WA –module to the sum Z[ΣA]+Z[Σ′], where Σ′ is the complement Σ ΣA . Because of the way in which \ the normalizer extension ν(WA) depends on the reflection extension ρ(WA), it is enough to check that the restriction of the reflection extension ρ(W ) to WA ′ is the sum of ρ(WA) and a semidirect product extension of WA by Z[Σ ]. As in Section 6.11, write Σ = Σi as a union of W –conjugacy classes. Each conjugacy class Σ can then be written as a union i `

Σi = j Σi,j ` of WA –conjugacy classes. For each i let τi denote a representative of the W – conjugacy class Σi , and τi,j a representative of the WA –conjugacy class Σi,j ; the centralizer of τi in W is Ci and the isotropy subgroup of τi,j in WA is Ci,j . The group C has a canonical extension by Z ((6.12)); if τ W then C is i i,j ∈ A i,j the centralizer of τi,j in WA and it too has a canonical extension by Z. Denote the corresponding extension classes by k H2(C ; Z) and k H2(C ; Z). i ∈ i i,j ∈ i,j Let ui : C W , ui,j : C W , and v : W W be the inclusion maps. i → i,j → A A → In the notation of Section 10.3, the task comes down to checking for each index i the validity of the formula

v∗ ui (k )= ui,j(k ) 0 , (10.6) # i ⊕τi,j ∈WA # i,j ⊕τi,j ∈/WA i,j 2 where 0i,j denotes the zero element of H (WA; Z[Σi,j]).

The left hand side of (10.6) can be evaluated with Lemma 10.5. The double cosets W W/C in question correspond to the orbits of the conjugation action A\ i of WA on Σi . For each such orbit Σi,j with orbit representative τ = τi,j , choose an element t W such that tτ t−1 = τ ; the element t is then the double coset ∈ i representative. Note that C = W tC t−1 . We have to consider the maps i,j A ∩ i

t−1(–)t v : C tC t−1 C i,j i,j → i −−−−→ i ui,j : C W i,j → A

i,j ∗ and compute u# vi,j(ki). The key observation is that if τ is not contained in WA , then Ci,j , which is generated by the reﬂections it contains [33, 1.5], ⊥ −1 ∗ must lie in tCi t ; hence vi,j(ki)=0. If τ is contained in WA , then Ci,j is the centralizer of τ in W , t−1τt = τ , and t−1C⊥ t C⊥ , so that v∗ (k ) = A i i,j ⊂ i i,j i ki,j .

eometry & opology, Volume 9 (2005) G T Normalizers of tori 1379

References

[1] K S Andersen, The normalizer splitting conjecture for p-compact groups, Fund. Math. 161 (1999) 1–16 MR1713198 [2] K S Andersen, J Grodal, J Møller, A Viruel, The classification of p– compact groups for p odd, Annals of Math. (to appear) arXiv:math.AT/0302346 [3] T Bauer, N Kitchloo, D Notbohm, E K Pedersen, Finite loop spaces are manifolds, Acta Math. 192 (2004) 5–31 [4] N Bourbaki, Eléments´ de mathématique. Fasc. XXXIV. Groupes et algèbres de Lie. Chapitre IV: Groupes de Coxeter et systèmes de Tits. Chapitre V: Groupes engendrés par des réflexions. Chapitre VI: systèmes de racines, Actualités Sci- entifiques et Industrielles 1337, Hermann, Paris (1968) MR0240238 [5] N Bourbaki, Eléments´ de mathématique: groupes et algèbres de Lie, chapitre 9: Groupes de Lie réels compacts, Masson, Paris (1982) MR0682756 [6] K S Brown, Cohomology of groups, Graduate Texts in Mathematics 87, Springer–Verlag, New York (1994) MR1324339 [7] A Clark, J Ewing, The realization of polynomial algebras as cohomology rings, Pacific J. Math. 50 (1974) 425–434 MR0367979 [8] M Curtis, A Wiederhold, B Williams, Normalizers of maximal tori, from: “Localization in group theory and homotopy theory, and related topics (Sympos. Battelle Seattle Res. Center, Seattle, Wash. 1974)”, Springer, Berlin (1974) 31– 47. Lecture Notes in Math. Vol. 418 MR0376956 [9] W G Dwyer, A Zabrodsky, Maps between classifying spaces, from: “Alge- braic topology, Barcelona, 1986”, Lecture Notes in Math. 1298, Springer, Berlin (1987) 106–119 MR928826 [10] W G Dwyer, C W Wilkerson, A new finite loop space at the prime two, J. Amer. Math. Soc. 6 (1993) 37–64 MR1161306 [11] W G Dwyer, C W Wilkerson, Homotopy fixed-point methods for Lie groups and finite loop spaces, Ann. of Math. (2) 139 (1994) 395–442 MR1274096 [12] W G Dwyer, C W Wilkerson, The center of a p-compact group, from: “The Cechˇ centennial (Boston, MA, 1993)”, Contemp. Math. 181, Amer. Math. Soc. Providence, RI (1995) 119–157 MR1320990 [13] W G Dwyer, C W Wilkerson, Product splittings for p-compact groups, Fund. Math. 147 (1995) 279–300 MR1348723 [14] W G Dwyer, C W Wilkerson, The elementary geometric structure of compact Lie groups, Bull. London Math. Soc. 30 (1998) 337–364 MR1620888 [15] W G Dwyer, CW Wilkerson, Centers and Coxeter elements, from: “Ho- motopy methods in algebraic topology (Boulder, CO, 1999)”, Contemp. Math. 271, Amer. Math. Soc. Providence, RI (2001) 53–75 MR1831347

eometry & opology, Volume 9 (2005) G T 1380 W G Dwyer and C W Wilkerson

[16] W G Dwyer, HR Miller, CW Wilkerson, The homotopic uniqueness of BS3 , from: “Algebraic topology, Barcelona, 1986”, Lecture Notes in Math. 1298, Springer, Berlin (1987) 90–105 MR928825 [17] M J Dyer, Hecke algebras and reflections in Coxeter groups, PhD thesis, Uni- versity of Sydney (1987) [18] S Helgason, Differential geometry and symmetric spaces, Pure and Applied Mathematics, Vol. XII, Academic Press, New York (1962) MR0145455 [19] J E Humphreys, Reflection groups and Coxeter groups, Cambridge Studies in Advanced Mathematics 29, Cambridge University Press, Cambridge (1990) MR1066460 [20] J Lannes, Théorie homotopique des groupes de Lie (d’après W G Dwyer etCW Wilkerson), Astérisque (1995) Exp. No. 776, 3, 21–45 MR1321642 [21] J Møller, D Notbohm, Centers and finite coverings of finite loop spaces, J. Reine Angew. Math. 456 (1994) 99–133 MR1301453 [22] J Møller, Homotopy Lie groups, Bull. Amer. Math. Soc. (N.S.) 32 (1995) 413– 428 MR1322786 [23] J Møller, N -determined p-compact groups, Fund. Math. 173 (2002) 201–300 MR1925483 [24] D Notbohm, On the 2–compact group DI(4), J. Reine Angew. Math. 555 (2003) 163–185 MR1956596 [25] D Notbohm, Maps between classifying spaces, Math. Z. 207 (1991) 153–168 MR1106820 [26] D Notbohm, Classifying spaces of compact Lie groups and finite loop spaces, from: “Handbook of algebraic topology”, North–Holland, Amsterdam (1995) 1049–1094 MR1361906 [27] D Notbohm, A uniqueness result for orthogonal groups as 2–compact groups, Arch. Math. (Basel) 78 (2002) 110–119 MR1888011 [28] D Notbohm, On the “classifying space” functor for compact Lie groups, J. London Math. Soc. (2) 52 (1995) 185–198 MR1345725 [29] A L Onishchik, E` B Vinberg, Lie groups and algebraic groups, Springer Series in Soviet Mathematics, Springer–Verlag, Berlin (1990) MR1064110 [30] A Osse, Sur le classifiant d’un groupe de Lie compact connexe, C. R. Acad. Sci. Paris Sér. I Math. 315 (1992) 833–838 MR1184913 [31] J-P Serre, Algèbres de Lie semi-simples complexes, WA Benjamin, Inc. New York–Amsterdam (1966) MR0215886 [32] T A Springer, Linear algebraic groups, Progress in Mathematics 9, Birkhäuser Boston Inc. Boston, MA (1998) MR1642713 [33] R Steinberg, Differential equations invariant under finite reflection groups, Trans. Amer. Math. Soc. 112 (1964) 392–400 MR0167535 [34] J Tits, Normalisateurs de tores. I. Groupes de Coxeter étendus, J. Algebra 4 (1966) 96–116 MR0206117

eometry & opology, Volume 9 (2005) G T